ArticlePDF Available

AI Health Agents: Pathway2vec, ReflectE, Category Theory, and Longevity

Authors:
  • DIYgenomics

Abstract

Health Agents are introduced as the concept of a personalized AI health advisor overlay for continuous health monitoring (e.g. 1000x/minute) medical-grade smartwatches and wearables for “healthcare by app” instead of “sickcare by appointment.” Individuals can customize the level of detail in the information they view. Health Agents “speak” natural language to humans and formal language to the computational infrastructure, possibly outputting the mathematics of personalized homeostatic health as part of their reinforcement learning agent behavior. As an AI health interface, the agent facilitates the management of precision medicine as a service. Healthy longevity is a high-profile area characterized by the increasing acceptance of medical intervention, longevity biotech venture capital investment, and global priority as 2 billion people will be over 65 in 2050. Aging hallmarks, biomarkers, and clocks provide a quantitative measure for intervention. Some of the leading interventions include metformin, rapamycin, spermidine, NAD+/sirtuins, alpha-ketoglutarate, and taurine. AI-driven digital biology, longevity medicine, and Web3 personalized healthcare come together in the idea of Health Agents. This Web3 genAI tool for automated health management, specifically via digital-biological twins and pathway2vec approaches, demonstrates human-AI intelligence amplification and works towards healthy longevity for global well-being.
AI Health Agents: Longevity, Pathway2vec, ReflectE, and Category Theory
Melanie Swan1, 2, Takashi Kido3, Eric Roland4, Renato P. dos Santos5
1DIYgenomics
2University College London
3Teikyo University
4RedBud AI, LLC
5Lutheran University of Brazil
melanie@DIYgenomics.org, kido.takashi@gmail.com, eric.roland@gmail.com, renatopsantos@ulbra.edu.br
Abstract
Health Agents are introduced as personalized AI health advi-
sors for “healthcare by app” instead of “sickcare by appoint-
ment,” especially to target Healthy Longevity as a global
wellness priority with two billion people estimated to be over
65 in 2050. Health Agents could allow physicians to oversee
thousands of patients simultaneously, addressing the 50% of
the world’s population still not covered by essential health
services. As AI Health Interfaces shift to continuous health
monitoring (1000x/minute) with medical-grade smart-
watches, pins, and wearables, individuals can customize the
level of information viewed. As any genAI agent system,
Health Agents “speak” natural language to humans and for-
mal language (as Math Agents) to the computational infra-
structure, possibly outputting the mathematics of personal-
ized homeostatic health as part of their reinforcement learn-
ing agent behavior. Health Agents could deliver precision
medicine as a service. Longevity may be achieved 80% with
sleep, diet, exercise, and stress reduction, and 20% by medi-
cal intervention (metformin, rapamycin, NAD+/sirtuins, al-
pha-ketoglutarate, taurine), measured quantitatively with ag-
ing clocks, biomarkers, and hallmarks. Health Agents are a
web3 genAI tool for automated health management, via Per-
sonalized Aging Clocks, digital-biological twins, and path-
way2vec approaches, for human-AI intelligence amplifica-
tion towards healthy longevity for global well-being.
The AI Longevity Mindset
The AI Mindset
The AI Stack. The AI infrastructure is evolving rapidly, par-
ticularly with genAI (generative AI which creates new data
based on what it has learned from a training dataset). Activ-
ity can be ordered in the four tiers of human-interface AI
assistants, reinforcement learning (RL) agents (self-driving,
robotics), knowledge graphs, and artificial neural network
architectures (ANNs). AI assistants and RL agents (embod-
ied through prompting) are an intelligence amplification
Copyright © 2024, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
tool for human-AI collaborative access to the vast range of
knowledge and computational resources now available.
ANNs. The first neural network architecture to deliver
genAI at scale is transformers (GPTs, generative pretrained
transformer neural networks), Large Language Models
(LLMs) which use attention as the mechanism to process all
connections in a dataset simultaneously to perform next
word (any token) prediction (OpenAI 2023). LLMs treat a
data corpus as a language, with syntax, semantics, and gram-
mar, whether natural language, mathematics, computer
code, or proteins. These kinds of Foundation Models are
trained on broad internet-scale data for application to a wide
range of use cases. Transformers are so-called because they
“transform” vector-based data representations during the
learning phase (using linear algebra methods).
Transformer architectures are being extended with state-
of-the-art LLMs released for multimodal VLMs (vision-lan-
guage models) (Gemini 2023), larger context windows (e.g.
genome-scale training, 1 million base pair size context win-
dow (HyenaDNA, Nguyen et al. 2023)), and longer sequen-
tial data processing with various convolutional and other
methods such as SSMs (structured state space models
(Mamba, Gu and Dao, 2023)) and model grafting (hybrid
network architectures evolving during training,
StripedHyena-7b (7 billion parameters (learned weights be-
tween data elements), Poli et al. 2023).
GPTs to GNNs: 2D to 3D+. An advance in digital biology
is GNNs (graph neural networks, technically a form of trans-
former) to process 3D data such as molecules (Bronstein et
al. 2021) with attention or message-passing. The early suc-
cess of GPTs is credited to the “traditional” machine learn-
ing recipe (Halevy et al. 2009) of a small set of algorithms
operating on a very large dataset, with substantial computa-
tional power. GNNs require a more extensive implementa-
tion of physics to treat 3D environments. The transfor-
mations of data representations in GNNs are more closely
AAAI Spring Symposium Series (SSS-24)
426
tied to the three main symmetry transformations in physics:
translation (displacement), rotation, and reflection, and the
notions of invariance (output unchanged per transformation)
and equivariance (output changes consistently with transfor-
mation). For example, AlphaFold2’s Invariant Point Atten-
tion models the displacement and rotation of amino acids as
triangles in space to identify pairwise combinations based
on angle and torsional force (Jumper et al. 2021). Also used
in GNNs is beyond-Euclidean hyperbolic space to effi-
ciently represent large datasets, for example hierarchical
tree-structured data (Zhou et al. 2022).
Knowledge Graph Embedding. Knowledge Graph vector
Embedding (KGE) methods also employ a full range of hy-
perbolic space and symmetry transformations, with found-
ing algorithms TransE (translation embedding) (Bordes et
al. 2013), RotatE (rotation embedding) (Sun et al. 2019), and
ReflectE (reflection embedding) (Zhang et al. 2022). More
capacious number systems expand from the everyday real
numbers (1D numbers) to 2D complex numbers with Com-
plEx (Trouillon et al. 2016) and 4D quaternion numbers
with QuatE (Zhang et al. 2019). Quantum formulations are
in development, e.g. quantum embedding (Li et al. 2023a)
and baqprop (quantum backpropagation of errors) (Verdon
et al. 2019). Temporal KGEs are a discovery domain with
temporal symmetry, antisymmetry, and inversion deployed
via Lorentz transformation (LorenTzE, Li et al. 2023b), ten-
sor factorization (TSimplE, He et al. 2023), and eGNN neu-
ral operator temporal dynamics (Xu et al. 2024). Finally,
KGE efforts are abstracted to mathematical formalism with
geometric algebras (GeomE, Xu et al. 2020), group-theo-
retic semigroups (SemE, Yang et al. 2022), and Riemannian
optimization (OrthogonalE, Zhu and Shimodaira, 2024).
The Formalization Turn. ANNs and KGE methods high-
light the implementation of mathematical physics in the AI
infrastructure, notably quantum-classical-relativistic mod-
els, real-complex-quaternionic (1D-2D-4D) numbers, and
beyond-Euclidean space (spherical, hyperbolic) and time
(Lorentz invariance, imaginary (complex-valued) time, and
time reversal symmetry). A second aspect of the “formali-
zation turn” continues the project of integrating disciplinary
fields by finding mathematical structure underlying them
(Wigner 1960). For example, combinatoric and geometric
structure in particle scattering amplitudes (Arkani-Hamed et
al. 2024), a category-theoretic account of double-entry
bookkeeping (Katis et al. 2008), and the formal axioms of
blockchains (Goncharov and Nechesov 2023).
Category theory is a meta-mathematics for investigating
the relationships between different types of mathematical
objects (e.g. sets, groups, vector spaces) using categories
and functions between them, composing their relations. Cat-
egory-theoretic methods are seen in deep learning (Gav-
ranovic et al. 2024), information theory (Katsumata et al.
2023), and genomics (Wu 2023), and generally proposed for
treating the complexity of biosystems (Rosen 1991).
Math Agents. Math Agents are an “AI math layer” for
various mathematical and computational tasks (Swan et al.
2023). Math is the data corpus processed with vector em-
bedding and visualized in equation clusters to view the
mathscape (set of equations) of a paper or sector at once.
The implication of embedding as a standard genAI method
is treating “big data” (entire data corpora) at the level of em-
bedding (a mathematical formulation) to deliver a clean ab-
stract view of very-large datasets. Embedding spaces allow
not only data viewing but novel discovery. For example, a
Universal Cell Embedding foundation model (representing
every cell state and type) was used to identify new develop-
mental lineages (Rosen et al. 2023), and a UMAP (compres-
sion) visualization of zero-shot embedding was used to find
novel mouse kidney cell types (Kragsteen et al. 2023).
GenAI means content generation, and math agent systems
may be prompted to write the mathematics of the underlying
knowledge graphs “for free” as part of their output. Not only
is the content-level prediction obtained (e.g. folded protein
structure), but also its mathematical description. AI is a
method for interacting with reality at the level of math (a
composite view of “big data” and “big math”).
Health Agents. Health Agents are envisioned as person-
alized AI health advisors for “healthcare by app” instead of
“sickcare by appointment.” Health Agents have two audi-
ences: human and AI in generating the content-level predic-
tions of personalized health and longevity interventions, to-
gether with the formal-level of mathematics describing ho-
meostatic health. Health agent systems could operate by
wearable apps from underlying blockchain-based healthcare
digital-biological twin platforms allowing physicians to
oversee thousands of patients simultaneously.
Digital Biology. Digital Biology is the extension of com-
putational biology informatics with genAI methods. Given
the complexity of biology, mathematics as a discovery tool
has been infeasible. It has not yet been possible to write the
robust mathematics of biology as formalizations explaining
pathology and homeostasis. However, Health Agents could
bring “mathematics as a method” to biological theorizing.
Agile Mindset. The AI mindset suggests a constant first-
principles stance in the era of Digital Biology. The molecu-
lar biology dogma of the DNA-RNA-protein synthesis chain
is being reversed from top-down to also include bottom-up
protein structure to DNA (e.g. in AlphaMissense (Cheng et
al. 2023)). Drug design is replacing drug discovery in the
idea of simply designing molecules with needed properties
instead of performing trial-and-error drug searches (Stokes
et al. 2020). Treating the pathway not the condition is a new
ethos in systems biology (Gschwind et al. 2023).
The Longevity Mindset
Another mindset shift is considering aging as a treatable dis-
ease instead of as a natural and inevitable condition of life.
427
The World Health Organization updated its International
Classification of Diseases (ICD-11) in 2022 for a diagnostic
category of “ageing associated decline in intrinsic capacity”
(Rabheru et al. 2022). Reducing suffering through longevity
therapies could be a worldwide priority given the “senior
tsunami” of 20% of the world’s population estimated to be
60 or older in 2050 (United Nations 2017). Longevity is de-
fined as a healthy lifespan of vitality, energy, and wellness,
contra aging as an exponential decline in capabilities leading
to age-related diseases and death. The aim is population-
scale interventions to slow, reverse, and prevent aging.
Longevity Clocks. One tool for measuring age and inter-
ventional impact is aging clocks which compare biological
age to chronological age. Such Personalized Aging Clocks
include a variety of epigenetic, transcriptomic, glycan,
metabolomic, and telomere length clocks at the organism
and organ level (Polidori 2024). Blood tests are used meas-
ure these factors, for example, the plasma protein signature
for eleven organ-specific aging clocks (brain, muscle, ar-
tery, heart, lung, immune, liver, kidney, pancreas, adipose,
intestine) to find 20% of 5,676 healthy adults already having
accelerated aging in at least one organ (Oh et al. 2023). Ep-
igenetic clocks confirmed the rejuvenation of six tissues get-
ting younger as measured by DNA methylation values (in
an animal parabiosis model) (Horvath et al. 2024).
Aging clocks may be used in concert with biomarkers of
aging (Moqri et al. 2023) and ageotypes (phenotypic age-
typing by metabolic, immune, liver, and kidney health)
(Ahadi et al. 2020) in targeted longevity interventions. The
aim is turning the biological clock back in 10-year periods
(e.g. a 70–80-year-old having the muscle health of a 60–70-
year-old), and then possibly maintaining people at a desired
biological age which may be 20-40 (Bischof et al. 2023).
Longevity Interventions. The experimental evidence for
longevity interventions continues to grow (Orr et al. 2024,
Blagosklonny 2023, Barzilai et al. 2016, Matysek et al.
2023, Soh et al. 2023, Fahy et al. 2019). Suggested geropro-
tective medications and supplements include rapamycin,
metformin, senolytics, acarbose, spermidine, NR/NAD+ en-
hancers, NSAIDs, lithium, glucosamine, glycine, and alpha-
ketoglutarate (Guarente et al. 2024, Gyanwali et al. 2022,
Partridge et al. 2020). The two leading interventions with
demonstrable results are rapamycin and metformin, in com-
bination activating AMPK and decreasing mTORC1 signal-
ing which may optimize the allocation of energy resources
towards the maintenance of proteostasis (protein homeosta-
sis) (Wolff et al. 2020).
Semaglutide Boom. Adding to aging clocks and aging bi-
omarkers as actionable approaches to longevity is the sur-
prise that 3% of Americans may already be taking an anti-
aging drug without knowing it. Named Sciences 2023
breakthrough of the year, semaglutide weight loss drugs
(GLP-1 agonists such as Wegovy, Ozempic) may also have
cardiac benefits and an anti-inflammatory role in the brain-
gut axis (Wong et al. 2023). Semaglutide is a medication
which mimics the GLP-1 (glucagon-like peptide-1) hor-
mone released in the gut to help the body feel full, producing
insulin and reducing blood sugar (glucose). The digital
health divide is a pressing concern as on the one hand, Wall
Street analysts estimate that worldwide spending on semag-
lutide, mostly not covered by insurance, could reach $100
billion by 2035 (Adegbesan 2023). On the other hand, the
World Health Organization notes that more than half of the
global population is still not covered by essential health ser-
vices (Taylor 2023). Implementing longevity protocols to
extend healthy lifespan could become an ethical imperative
and a matter of equity, access, and business model.
Longevity – There’s an app for that~! The longevity rev-
olution could be by app – implemented with Health Agent
wearables, sensors, patches, apps, and 3D printers, moni-
tored by longevity physicians, with digital twin partners
(virtual patient simulations). Wearables capturing tempera-
ture, sleep quality, and heart rate variability provide address-
able early-warning signs for various pathologies (Alavi et
al. 2022), for example sleep quality predicting type 2 diabe-
tes onset by ten years (Komine et al. 2016). Technology-
savvy populations suggest an uberized (widespread accessi-
ble via mobile technology) approach to healthcare and lon-
gevity therapy delivery, as Deloitte confirms 90% world-
wide mobile phone penetration in 2017 (smartphones 80%;
81% in emerging markets) (Wigginton et al. 2017).
Forward-looking countries are targeting longevity as a
government policy initiative with goals for healthy citizenry
with +5-year healthspans in Singapore (the sixth “Blue
Zone” country) and Arab states (Kalin 2023). Health is
emerging as a competitive currency and basic human right
for human potentiality (Nussbaum 2003). The XPRIZE
Longevity Prize was announced in November 2023 for a
therapeutic intervention to restore muscle, cognitive, and
immune function 10-20 years in 65–80-year-old populations
within one year. Longevity venture capital, although down
from 2021 peaks, reported 101 deals and USD $1.1 billion
for the first three quarters of 2023, sector tracker Longev-
ity.Technology reported (Newman and Belleza 2023) and
eleven dedicated Longevity venture funds were profiled by
Forbes (Predin 2023). AI for social good” has important
uses in improving health outcomes (Tomasav et al. 2020).
Generative AI and Biology
Natural language is the first area of demonstrable genAI pro-
gress, however, biology may be orders of magnitude more
complex including because the “ruleset” is unknown.
Whereas one of the largest open-source foundation models,
LLaMA, has 65 billion parameters (learnable weights be-
tween entities) (MetaAI 2023), state-of-the-art protein mod-
els have 100 billion parameters (Chen et al. 2023), and
428
genome language models may require even more. Biologi-
cal computational complexity classes (protein, genome,
pathway) could be formalized from earlier graph visualiza-
tion starting points (Cirillo et al. 2018, Kugler et al. 2010).
Protein Language Models. Life science AI foundation
models include BioMap’s xTrimoPGLM (cross-modal in-
teractome and multiomics transformer) with 100 billion pa-
rameters as mentioned (Chen et al. 2023) and MetaAI’s
ESM-2(15B) (evolutionary stochastic model) with 15 bil-
lion parameters (Lin et al. 2022). They represent the protein
language internally as opposed to beginning with evolution-
ary MSA (multiple sequence alignment) reads as in Al-
phaFold2 and RoseTTAFold. Digital biology platforms
such as NVDIA’s BioNeMo (biological neural modeling)
offer pretrained drug discovery models as a cloud service.
Genome Language Models. AI foundation models are be-
ing developed in genomics for sequencing and analysis. In
sequencing, DeepVariant is a Stanford-led transformer pro-
ject that holds the Guiness Book of World Records for the
fastest human genome sequenced (5 hours 2 min, on 16 Mar
2021, still unbeaten as of February 2024). The DeepConsen-
sus project uses a gap-aware sequence transformer to reduce
read errors by 42% as compared with hidden Markov mod-
els as the traditional sequence-reading method (Baid et al.
2023). DNAGPT is a transformer performing sequence clas-
sification through numerical regression and a comprehen-
sive token language (Zhang et al. 2023).
In genomic analysis, there are various projects focused on
building genome-scale language models such as Hye-
naDNA, training on whole-genome datasets to model indi-
vidual mutations (Nguyen et al. 2023). The idea is to be able
to prompt ChatGPT with an entire human genome to iden-
tify mutational profile risks and interventions (e.g. which
aging clocks to start with in precision health programs).
Other projects include DNABERT for making predictions
about transcription factor binding sites, scBERT trained on
scRNA-Seq (single cell RNA sequencing) data to predict
gene-gene interactions, and Enformer for making predic-
tions about long-range interactions in the genome. AlphaM-
issense uses protein structures to predict pathogenic mis-
sense mutations (of 71 million human mutations, 32% are
pathogenic) (Cheng et al. 2023). Missense mutations are
~58% of mutations; then nonsense (10%), frameshift (8%),
splice (6%), insertion-deletion (5%), and other (13%).
Digital Twins and Biology
A digital twin is a virtual representation of a physical object,
person, or process, estimated by McKinsey to be a $48.2 bil-
lion industry in 2026 (Borden 2023). Digital twins are used
to model manufacturing operations and infrastructure. Sin-
gapore completed the first digital twin of an entire country
(Virtual Singapore) in 2022. In healthcare, virtual patient
models are used for procedure simulation, medical educa-
tion, clinical research, and drug development. Clinical trial
simulations of millions of genAI-created virtual patients
might be routine in the future.
On the one hand, a long-term vision supports the idea of
there being biomedical digital twins for the world’s 8.1 bil-
lion humans. Each person could have an ID number, a MAC
address (phone), and a secure web3 digital twin (Akash and
Ferdous 2022). On the other hand, healthcare digital twins
are not an immediate possibility given the complexity of bi-
ological systems, the need for large-scale, high-quality data,
and the potential for model inaccuracies.
Longevity Twins. Homeostatic health could be Turing-
complete (a format running on any platform, biological or
machine; digital twin or biological counterpart). Digital
twins started for longevity medicine could extend to future
BCI (brain-computer interface) and connectome projects.
Web3
Web3 refers to the current third phase of the internet’s de-
velopment in expanding the interaction mode from the pas-
sive read-only web (1990s) to the interactive read-write web
(2000s) to the secure and remunerative read-write-own web
(2020s enabled by web3 blockchain technologies). Digital
transformation continues as many industries become in-
creasingly digitally instantiated. First was the “ready” con-
version of content (1990s dot-com news, media, and enter-
tainment), then followed by the more complicated imple-
mentation of money and economics, digital art and intellec-
tual property (IP), supply chain, manufacturing, transporta-
tion, and science. These latter require more complex fea-
tures such as non-fungibility and contracts, capabilities pro-
vided by blockchains. Blockchains are secure distributed
ledger systems providing a database for resource allocation
and an immutable record of event histories. Blockchain eco-
systems are a foundational information technology using se-
cure properties for digital exchange and economics more
broadly as a design principle to produce non-economic out-
comes with a broader society benefit (Swan 2015).
Blockchains in Biology
Blockchains are used in health and biology for secure data
transfer, supply chain logistics, chain-of-custody tracing,
and clinical trials. MediLedger is a global pharmaceutical
supply chain blockchain (led by Pfizer, Amgen, and Gilead)
completing a pilot program with the U.S. FDA in 2023 to-
wards the 2023 Drug Supply Chain Security Act. Triall is a
clinical trials blockchain platform conducting a two-year
multi-center pulmonary arterial hypertension clinical trial
with the Mayo Clinic. BloodChain is a blood donation net-
work managed with blockchains.
429
Genome Blockchains. The scale of contemporary science
(million-patient studies involving 30 GB whole-human ge-
nome files from sixteen sites (Bellenguez et al. 2022)) im-
plies new models for its conduct. For example, the direct-to-
consumer whole-human genome sequencing company Neb-
ula Genomics mints a new user-controlled NFT for each se-
quenced genome (with genomes.io).
DeSci and Longevity DAOs. Web3 methods facilitate
DeSci (decentralized science) for carrying out internet-scale
bioinformatics. Example projects include LabDAO (an open
science collaboration community) and VitaDAO (specific to
longevity research). (A DAO is a distributed autonomous
organization, an entity formed with blockchain-based smart
contracts, with some level of automated administration.)
Blockchain Healthcare Digital Twins
The healthcare delivery system of the future could be one
orchestrated by health agents and healthcare digital twins, as
physicians oversee the smart health ecosystem by app.
Blockchains are suggested as a platform for healthcare dig-
ital twins for several reasons. First is the usual notion of
blockchains for secure multi-party access to a single un-
changeable event history. Second is proof attestation to
track the efficacy of interventions (insurance companies are
already considering using aging clocks in actuarial tables).
Third is an interoperable overlay for integrating multiple
omics data streams and EHRs. Fourth is blockchain design
principles for modeling non-economic aspects such as ho-
meostasis as its own “bioeconomy.” Fifth is the ability to
add genAI technologies to the secure health stack with
Health Agents as the “user” of the blockchain healthcare
digital twin; blockchains both track and facilitate the de-
ployment of AI agents. Health Agents could help the digital
twin learn its own longevity protocol.
Blockchain Healthcare Digital Twins for Longevity. The
complexity of pathways and processes in the human body
can be modeled in the blockchain healthcare digital twin as
a multiscalar homeostatic economy of wellness and disease.
A blockchain system can instantiate the body, labeling enti-
ties as wallet addresses, giving them relevant biocurrency
balances, and modeling their activity with smart contract
transactions. The schema allows any level of drill-down and
roll-up for views of the system per the hashing structure (e.g.
organ, tissue, cellular level). One top-level Merkle root can
call the entirety of the body. One lowest-level transaction
could record the amount of insulin-facilitated glucose re-
lease into a cell. The complex pathways of the systems biol-
ogy of aging (Furber 2019) and their related interventions
could be modeled with a blockchain smart contract system.
As Virtual Singapore’s first digital twin of a country, the
first full digital twin of the human can be imagined. The
eleven-organ aging clocks could have avatorial representa-
tion sitting around “the conference table of the body” as the
user interface, genAI stating their agenda based on real-life
biomarker levels (analogous to other non-human entities AI-
voiced through sensor output).
Longevity State Machine. A smart contract system is im-
plicated to automate blockchain healthcare digital twins.
This could be via the Polkadot blockchain ecosystem. Vari-
ous chains federate for interoperability in the overall struc-
ture of “block space as a service” (secure computation and
event-recording). Proof-of-stake (randomized participant-
based) mining systems provide a greener alternative to
costly Bitcoin proof-of-work mining.
Smart contract state machines update wallet balances per
transactions. Wallets may contain any cargo such as crypto-
currencies, NFTs (digital assets), computer memory, iden-
tity documents (digital passport), or biocurrencies used by
Health Agents to manage longevity programs in healthcare
digital twins. Independent observers (oracles) take readings
(e.g. from wearables, blood tests, patches, toilets, and apps),
sending attested measurements to the smart contract system.
Biocurrency Transactions. In a biosystem, there are many
different “biocurrencies” circulating to conduct homeostatic
activities. The longevity twin has wallet addresses for hun-
dreds of blood biomarkers (e.g. glucose, insulin, homocys-
teine, HS-CRP, Hb1ac, lymphocytes); everything seen on a
blood test. For example, a homocysteine wallet might have
an initial balance of 9.0, which is too high, as the desired
level of the biomarker may be 5.0. The target protocol is
written as a smart contract with logic about the course of
action, for example, reducing homocysteine with folate.
Each intervention, supplement, or prescription drug could
have a wallet address and balance that is managed with
smart contracts. Given the homocysteine wallet balance of
9.0, the instruction is for a daily regimen of 500 mg of folate.
The folate wallet is activated with a transaction to dispense
500 mg into the smart smoothie, decrementing the overall
monthly balance of 50,000 to 49,500. Wallet entities may
have their own biocurrencies (e.g. homocysteine, folate), all
convertible to BioCoin or some other universal currency.
Biowallets and Longevity Payment Channels. Health
Agent smart contracts could manage longevity interventions
in off-chain payment channels (contractual interaction se-
quences) for daily interactions settled to the main health
chain on a weekly or monthly basis. Smart contracts would
obtain measures of pathway biocurrencies from connected
sensors (oracles) and distribute intervention-currencies as a
result. The payment channel orchestrates the folate balance
acting on the homocysteine pathway to reduce the homocys-
teine balance. HomocysteineCoin and FolateCoin are read-
ily convertible to meta-tokens LongevityCoin and BioCoin.
Bioconsensus. Health Agents can finetune the different
biocurrencies with payment channel smart contracts to learn
the optimal personalized longevity protocol. Ideally the sys-
tem can self-learn its homeostasis. Smart contract coordi-
nated biowallets could be rewarded by Health Agent
430
reinforcement learning mechanisms to reach their own bio-
consensus as to the truth state constituting optimal longev-
ity. The biowallet is a useful binary mechanism to target
positive and negative behavior, for example reversing dis-
ease as “bad player behavior” by assigning negative wallet
balances to missense mutations and transposon activations
that are targeted by the “good actor behavior” of epigenetic
methylation to earn token and pay down negative wallet bal-
ances. Health Agent blockchain longevity twins could be
tested with Lightning Network payment channels.
Aging Clocks Registered at Birth. As biological processes
peak and decline at various stages, some even before birth,
a full life-cycle longevity program could be used to register
aging clocks as wallets at birth. As life proceeds, balances
can be updated to compare biological and chronological age
and apply interventions. Blockchain healthcare digital twins
are further implicated to manage the potential future suite of
wearables, patches, and on-board electronics of brain-com-
puter interfaces, connectome maps, and medical nanorobots.
Security is crucial as 5 million people have implanted pace-
makers, some internet-connected for real-time remote mon-
itoring, automated pacing therapies, and software updates.
Health Agents
Health Agents are the concept of an AI system tasked with
precision human health and longevity. Such a personalized
health system app-tool entails dialogue and data capture at
the human-AI interface level and agent-based activity at the
AI infrastructure level to access information, model health,
and orchestrate intervention. Health Agents are qualitative
and quantitative, interacting at both the human-consumable
content-level of personalized health and longevity interven-
tion, and the AI-usable formal-level of mathematics describ-
ing homeostatic health (possibly involving category theory
and other formalisms). A digital twin system is implicated
as a virtual model of individual health, secured by block-
chain smart contracts, user-permissioned for agents to learn
by sharing federated data (as in vehicular blockchains of
self-driving networks) and for health care plan or societal
level aggregation. For precision longevity, the aim is to have
a formal statement of health (homeostasis) which can be
learned by AI Health Agent systems. The results could be
connected to aging clock, biometrics, and tracking data for
longevity physicians to review in the Longevity App.
Pathway2vec
Aging clocks research has identified pathways not tradition-
ally targeted by longevity interventions (Oh et al. 2023),
hence the first project for Health Agents is implementing
Personalized Aging Clocks in a pathway2vec program.
Pathway2vec is representing biological pathways as vectors
for input to ANNs, in a precedence of approaches beginning
with word2vec (Mikolov et al. 2013), and including
math2vec (vector representations of equations), gene2vec,
SNP2vec, mut2vec, and disease2vec. Related “cancer2vec”
approaches use cluster embedding to identify individual
cancer risk (Choy et al. 2019) and patient-specific molecular
patterns of cancer (Pfeifer et al. 2022), which could likewise
model personalized aging-clock targets in longevity. A path-
way2vec project defines a vector representation of drug
pairs and pathway genes (Yamagiwa et al. 2023).
Longevity Protocol Testing
The next steps in the realization of precision longevity med-
icine could be testing and tailoring proposed protocols for
personalized intervention (e.g. Partridge et al. 2020 p. 516,
Houston 2018 pp. 86-87, and Sinclair 2019 p. 304). This en-
tails blood biomarker tests to assess an individual’s ageo-
type profile of accelerated aging using eleven-organ system
aging clocks (Oh et al. 2023) and other aging clocks (e.g.
epigenetic) (Horvath et al. 2024).
Risks and Limitations
The biggest risk in the AI Health Agents proposal is that alt-
hough technology development is proceeding quickly, the
implementation of healthy longevity programs may be too
late to adequately respond to aging demographics. Longev-
ity scholars therefore promote the “escape velocity” idea of
deploying immediate interventions to buy enough time until
more complete solutions are available (de Grey 2004). Alt-
hough a contemporary lens sees longevity as a technology,
it may not be easy to reprogram biology. There are social
challenges to traditional ideas of health, life, and care, along
with regulatory hurdles and healthcare system adoption con-
straints. Further, the use of genAI in biology and medicine
has a new set of risks related to data quality, interpretability,
and ethics (hallucinations and bias), which requires attention
before Health Agents are deployed.
Conclusion
The build-out of the AI infrastructure is proceeding quickly,
particularly to facilitate the study of biology. Biocomplexity
poses a formidable challenge, but AI methods have encour-
aging results in the third wave of digitization. The goal is to
harness the new tools of digital biology to access a larger
scope of investigation with synthesis between areas and ac-
tionable steps. This work introduces the concept of Health
Agents (personal AI health stewards realized with block-
chain digital health twins) as a new idea in human-AI col-
laboration and intelligence amplification. Health Agents
provide an interface for not only qualitative knowledge
431
access, but more importantly “non math speaker” access to
the growing intensity of formal methods in the computa-
tional infrastructure. The immediate application of Health
Agents is healthy longevity, starting to be conceived as a
technology, readily implementable and optimizable like any
other.
Healthy Longevity and Society. Longevity medicine aims
to extend the healthy lifespan of humans by preventing and
treating age-related diseases. There are many potential ben-
efits for society. First is improving the quality of life and
well-being of older adults and families by reducing the bur-
den of chronic diseases, disability, and dependency (Fried et
al. 2022). Second is enhancing the productivity and contri-
bution of older adults to the economy and society, by ena-
bling people to work longer, volunteer, mentor, and partici-
pate in civic activities (Accius et al. 2022). Third is reducing
the healthcare costs associated with aging, by shifting the
focus from treating diseases to promoting health and pre-
venting disease (OECD 2019). Fourth is creating new op-
portunities for various industries, such as biotechnology,
pharmaceuticals, education, and entertainment that cater to
the needs of older adults (Ng and Indran 2024). Longevity
medicine realized with AI tools such as Health Agents is a
mindset and health program that could help individuals and
society to realize a longer, healthier, and happier existence.
References
Accius, J.; Ladner, J.; and Alexander, S. 2022. The Global Lon-
gevity Economy. AARP.
Adegbesan, A. 2023. Obesity Drug Market Set to Hit $100 Billion
by 2035, BMO Says. Bloomberg News.
Ahadi, S.; Zhou, W.; Rose, S. M. S.; et al. 2020. Personal aging
markers and ageotypes. Nat Med 26(1): 83–90.
Akash, S. S.; and Ferdous, M. S. 2022. A Blockchain Based Sys-
tem for Healthcare Digital Twin. IEEE Access 10:50523–50547.
Alavi, A.; Bogu, G. K.; Wang, M.; et al. 2022. Real-time alerting.
Nat Med 28(1): 175–184.
Arkani-Hamed, N.; Cao, Q.; Dong, J.; et al. 2024. NLSM Tr(φ
3). arXiv preprint arXiv:2401.05483.
Baid, G.; Cook, D. E.; Shafin, K.; et al. 2023. DeepConsensus. Nat
Biotechn 41(2): 232–238.
Basher, A.; and Hallam, S. J. 2021. Leveraging heterogeneous net-
work embedding for metabolic pathway prediction. Bioinform
37(6): 822–829.
Bellenguez, C.; Kucukali, F.; Jansen, I. E.; et al. 2022. New in-
sights into the genetic etiology of Alzheimer’s disease. Nat Genet
54(4): 412–436.
Blagosklonny, M. V. 2023. Towards disease-oriented dosing of ra-
pamycin for longevity Aging 15(14): 6632–6640.
Barzilai, N.; Crandall, J. P.; Kritchevsky, S. B.; et al. 2016. Met-
formin as a Tool to Target Aging. Cell Metab 23(6): 1060–1065.
Bischof, E.; Zhavoronkov, A.; Moskalev, A.; et al. 2023. Longev-
ity Medicine 101. Lecture 9.
Borden, K. 2023. What is digital-twin technology? McKinsey.
Bordes, A.; Usunier, N.; Garcia-Duran, A.; et al. 2013. Translating
embeddings for modeling MM data. NeurIPS 2787–2795.
Bronstein, M. M.; Bruna, J.; Cohen, T.; and Velickovic, P. 2021.
Geometric deep learning: Grids, groups, graphs, geodesics, and
gauges. arXiv preprint arXiv:2104.13478.
Chen, B.; Cheng, X.; Geng, Y.; et al. 2023. xTrimoPGLM: Unified
100B-Scale Pre-trained Transformer for Deciphering the Lan-
guage of Protein. bioRxiv doi: 10.1101/2023.07.05.547496v1
Cheng, J.; Novati, G.; Pan, J.; et al. 2023. Accurate proteome-wide
missense variant effect prediction with AlphaMissense. Science
381(6664): eadg7492.
Choy, C. T.; Wong, C. H.; and Chan, S. L. 2019. Infer related genes
from large scale gene expression. Front Genet Sec CG 9:682.
Cirillo, D.; Ponce-de-Leon, M.; and Valencia, A. 2018. Algorith-
mic complexity in computational biology: basics, challenges and
limitations. arXiv preprint arXiv:1811.07312.
Couzin-Frankel, J. 2023. Obesity meets its match. Science.
de Grey, A. D. N. J. 2004. Escape Velocity: Why the Prospect of
Extreme Human Life Extension Matters. PLoS Biol 2(6): e187.
Fahy, G. M.; Brooke, R. T.; Watson, J. P.; et al. 2019. Reversal of
epigenetic aging and immunosenescent trends in humans. Aging
Cell 18(6): e13028.
Fried, L. P.; Wong, J. E. L.; and Dzau, V. 2022. A global roadmap
for healthy longevity. Nat Aging 2: 1080–1083.
Furber, J. D. 2019. Systems Biology of Human Aging – Network
Model. Innov Aging 3(Suppl 1): S973.
Gavranovic, B.; Lessard, P.; Dudzik, A.; et al. 2024. Categorical
Deep Learning. arXiv preprint arXiv:2402.15332.
Gemini Team, Google. 2023. Gemini: A Family of Highly Capable
Multimodal Models. arXiv preprint arXiv:2312.11805v1.
Goncharov, S.; and Nechesov, A. 2023. Axiomatization of BC
Theory. Math 11: 2966.
Gschwind, A. R.; Mualim, K. S.; Karbalayghareh, A.; et al. 2023.
An encyclopedia of enhancer-gene regulatory interactions in the
human genome. bioRxiv doi: 10.1101/2023.11.09.563812.
Gu, A.; and Dao, T. 2023. Mamba: Linear-Time Sequence Model-
ing with Selective State Spaces. arXiv preprint arXiv:2312.00752.
Guarente, L.; Sinclair, D. A.; and Kroemer, G. 2024. Human trials
exploring anti-aging medicines. Cell Metab 36(2): 354-376.
Gyanwali, B.; Lim, Z. X.; Soh, J.; et al. 2022. Alpha-Ketoglutarate
dietary supplementation to improve health in humans. Trends En-
docrinol Metab 33(2): 136–146.
Halevy, A.; Norvig, P.; and Pereira, F. 2009. The unreasonable ef-
fectiveness of data. IEEE intelligent systems 24 (2): 8–12.
He, P.; Zhou, G.; Zhang, M.; et al. 2023. Improving temporal
knowledge graph embedding using tensor factorization. Appl Intell
53:8746–8760.
Houston, M. 2018. The role of noninvasive cardiovascular testing.
Ther Adv Cardiovasc Dis 12(3): 85–108.
Horvath, S.; Singh, K.; Raj, K.; et al. 2024. Reversal of biological
age in multiple rat organs by young porcine plasma fraction. Gero-
science 46(1): 367–394.
Ioannidis, V. N.; Song, X.; Manchanda, S.; et al. 2020. DRKG:
Drug Repurposing Knowledge Graph for Covid-19.
Jumper, J.; Evans, R.; Pritzel, A.; et al. 2021. Highly accurate pro-
tein structure prediction with AlphaFold. Nature 596: 583–89.
Kalin, S. 2023. Saudi Arabia Research on Aging. WSJ.
432
Katis, P.; Sabadini, N.; and Walters, R. F. C. 2008. On partita dop-
pia. arXiv preprint arXiv:0803.2429v1.
Katsumata, S.; Rival, X. and Dubut, J. 2023. A Categorical Frame-
work for Program Semantics. arXiv preprint arXiv:2309.08822.
Komine, T.; Takadama, K.; and Nishino, S. 2016. Toward Next-
Generation Sleep Monitoring. AAAI Spring Symposia 2016: Well-
Being Computing: AI Meets Health and Happiness Science.
Kragesteen, B. K.; Giladi, A.; David, E.; et al. 2023. The transcrip-
tional identity of erythropoietin cells. Nat Med 29: 1191–1200.
Kugler, H.; Larjo, A.; and Harel, D. 2010 Biocharts: a visual for-
malism for complex biological systems. JRSI 7(48):1015–1024.
Li, L.; Zhang, X.; Jin, Z.; et al. 2023a. KG completion: quantum
embedding and quaternion interaction. Info Sci 648(119548).
Li, M. M.; Huang, K.; and Zitnik, M. 2022. Graph representation
learning in biomedicine and healthcare. Nat Bio Eng 6: 1353–69.
Li, N.; Haihong, E.; Shi, L.; et al. 2023b. LorenTzE: Temporal
KGE. ICANN pp. 472–484.
Matysek, A.; Sun, L.; Kimmantudawage, S. P.; et al. 2023. Target-
ing impaired nutrient sensing via the sirtuin pathway. Ageing Res
Rev 90:102029.
MetaAI. 2023. LLaMA: A foundational, 65-bn-parameter LLM.
https://ai.meta.com/blog/large-language-model-llama-meta-ai/.
Mikolov, T.; Sutskever, I.; Chen, K.; et al. 2013. Distributed Rep-
resentations: Words and Phrases. arXiv preprint arXiv:1310.4546.
Moqri, M.; Herzog, C.; Poganik, J. R.; et at. 2023. Biomarkers of
aging for the identification and evaluation of longevity interven-
tions. Cell 186(18): 3758–3775.
Newman, P.; and Belleza, C. 2023. Q3 2023: Longevity Invest-
ment Report. Longevity.Technology.
Ng, R.; and Indran, N. 2024. Innovations for an Aging Society.
Gerontologist 64(2):gnad015.
Nguyen, E., Poli, M., Faizi, F.; et al. 2023. HyenaDNA: Long-
range genomic sequence. arXiv preprint arXiv:2306.15794v1.
Nussbaum, M. C. 2003. Capabilities as Fundamental Entitlements:
Sen and Social Justice. Feminist Economics 9(2-3): 33–59.
OECD. 2019. The Heavy Burden of Obesity. oe.cd/obesity2019.
Oh, H. S. H., Rutledge, J., Nachun, D.; et al. 2023. Organ aging
signatures in the plasma proteome track health and disease. Nature
624, 164–172.
OpenAI. 2023. GPT-4 Technical Report. arXiv preprint
arXiv:2303.08774v4.
Orr, M. E.; Kotkowski, E.; Ramirez, P.; et al. 2024. A randomized
placebo‑controlled trial of nicotinamide riboside. Geroscience
46:665–682.
Partridge, L.; Fuentealba, M.; and Kennedy, B. K. 2020. The quest
to slow ageing through drug discovery. Nat Rev Drug Discov
19(8): 513–532.
Pfeifer, B.; Saranti, A.; and Holzinger, A. 2022. GNN-SubNet: dis-
ease subnetwork explainable GNNs. Bioinform 38: ii120-ii126.
Poli, M.; Massaroli, S.; Nguyen, E.; et al. 2023. Hyena Hierarchy:
Larger Convolutional LMs. arXiv preprint arXiv:2302.10866.
Polidori, M. C. 2024. Aging hallmarks, biomarkers, and clocks for
personalized medicine. Free Radical Biol and Med 215: 48-55.
Predin, J. M. 2023. Venture Capital Giants in The Billion-Dollar
Quest for Longevity Breakthroughs. Forbes.
Rabheru, K.; Byles, J. E.; and Kalache, A. 2022. How “old age”
was withdrawn from ICD-11. The Lancet 3(7): e457–e459.
Rosen, R. 1991. Life itself: a comprehensive inquiry into the na-
ture, origin, and fabrication of life. New York: Columbia UP.
Rosen, Y.; Roohani, Y.; Agarwal, A.; et al. 2023. Universal Cell
Embeddings: A Foundation Model for Cell Biology.
Sinclair, D. A. 2019. Why We Age. New York: Atria.
Soh, J.; Raventhiran, S.; Lee, J. H.; et al. 2023. The effect of gly-
cine administration on the characteristics of physiological systems.
Geroscience 46, 219–239.
Stokes, J. M.; Yang, K.; Swanson, K. 2020. Deep Learning Ap-
proach to Antibiotic Discovery. Cell 180(4): 688-702.e13.
Sun, Z.; Deng, Z.; Nie, J.; and Tang, J. 2019. RotatE: KGE Rota-
tion in Complex Space. arXiv preprint arXiv:1902.10197.
Swan, M. 2015. Blockchain: Blueprint for a New Economy. Sebas-
topol CA: O’Reilly Media.
Swan, M.; Kido, T.; Roland, E.; and dos Santos, R. P. 2023. Math
Agents. arXiv preprint arXiv:2307.02502.
Taylor, L. 2023. More than half the world’s population lack basic
health services as progress stalls. BMJ 382: 2160.
Tomasev, N.; Cornebise, J.; Hutter, F.; et al. 2020. AI for social
good. Nat Commun 11: 2468.
Trouillon, T.; Welbl, J.; Riedel, S.; et al. 2016. Complex embed-
dings for simple link prediction. arXiv preprint arXiv:1606.06357.
United Nations. 2017. World population projected to reach 9.8 bil-
lion in 2050, and 11.2 billion in 2100. The 2017 Revision.
Verdon, G.; McCourt, T.; and X. 2019. Quantum Graph Neural
Networks. arXiv preprint arXiv:1909.12264.
Wigginton, C.; Curran, M.; and Brodeur, C. 2017. Global mobile
consumer trends, 2nd edition. Deloitte.
Wigner, E. P. 1960. The unreasonable effectiveness of mathemat-
ics in the natural sciences. Commun Pure Appl Math 13(1): 1–14.
Wolff, C. A.; Reid, J. J.; Musci, R. V.; et al. 2020. Differential Ef-
fects of Rapamycin and Metformin. J Gerontol 75(1): 32–39.
Wong, C. K.; McLean, B. A.; Baggio, L. L.; et al. 2023. Central
glucagon-like peptide 1. Cell Metab S1550–4131(23): 00420–5.
Wu, Y. 2023. A Category of Genes. arXiv preprint
arXiv:2311.08546v1.
Xu, C.; Nayyeri, M.; Chen, Y.-Y.; and Lehmann, J. 2020. KGE in
Geometric Algebras. arXiv preprint arXiv:2010.00989.
Xu, M.; Han, J.; Lou, A.; et al. 2024. Equivariant Graph Neural
Operator for 3D Dynamics. arXiv preprint arXiv:2401.11037v1.
Yamagiwa, H.; Hashimoto, R.; Arakane, K.; et al. 2023. Analogy
Tasks in BioConceptVec Biological Pathways. IEICE 123(91).
Yang, T.; Wang, Y.; Sha, L.; et al. 2022. Knowledgebra. ML and
Knowl Extr 4(2): 432–445.
Zhang, D.; Zhang, W.; and Zhao, U. 2023. DNAGPT: A General-
ized Pre-trained Tool for DNA Sequence Analysis. arXiv preprint
arXiv:2307.05628.
Zhang, Q.; Wang, R.; Yang, J.; and Xue, L. 2022. KGE Reflection
transformation. Knowledge-Based Systems 238 (107861).
Zhang, S.; Tay, Y.; Yao, L.; and Liu, Q. 2019. Quaternion
Knowledge Graph Embeddings. arXiv preprint arXiv:1904.10281.
Zhou, H.; Wang, W.; Jin, J; et al. 2022. GNN for PPIs. Molecules
27(18): 6135.
Zhu, Y.; and Shimodaira, H. (2024). Block-Diagonal Orthogonal
Relation for KGE. arXiv preprint arXiv:2401.05967.
433
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Animal studies show aging varies between individuals as well as between organs within an individual1–4, but whether this is true in humans and its effect on age-related diseases is unknown. We utilized levels of human blood plasma proteins originating from specific organs to measure organ-specific aging differences in living individuals. Using machine learning models, we analysed aging in 11 major organs and estimated organ age reproducibly in five independent cohorts encompassing 5,676 adults across the human lifespan. We discovered nearly 20% of the population show strongly accelerated age in one organ and 1.7% are multi-organ agers. Accelerated organ aging confers 20–50% higher mortality risk, and organ-specific diseases relate to faster aging of those organs. We find individuals with accelerated heart aging have a 250% increased heart failure risk and accelerated brain and vascular aging predict Alzheimer’s disease (AD) progression independently from and as strongly as plasma pTau-181 (ref. ⁵), the current best blood-based biomarker for AD. Our models link vascular calcification, extracellular matrix alterations and synaptic protein shedding to early cognitive decline. We introduce a simple and interpretable method to study organ aging using plasma proteomics data, predicting diseases and aging effects.
Article
Full-text available
Categorical semantics of type theories are often characterized as structure-preserving functors. This is because in category theory both the syntax and the domain of interpretation are uniformly treated as structured categories, so that we can express interpretations as structure-preserving functors between them. This mathematical characterization of semantics makes it convenient to manipulate and to reason about relationships between interpretations. Motivated by this success of functorial semantics, we address the question of finding a functorial analogue in abstract interpretation, a general framework for comparing semantics, so that we can bring similar benefits of functorial semantics to semantic abstractions used in abstract interpretation. Major differences concern the notion of interpretation that is being considered. Indeed, conventional semantics are value-based whereas abstract interpretation typically deals with more complex properties. In this paper, we propose a functorial approach to abstract interpretation and study associated fundamental concepts therein. In our approach, interpretations are expressed as oplax functors in the category of posets, and abstraction relations between interpretations are expressed as lax natural transformations representing concretizations. We present examples of these formal concepts from monadic semantics of programming languages and discuss soundness. Comment: MFPS 2023
Article
Full-text available
Nicotinamide riboside (NR) increases blood levels of NAD+, a cofactor central to energy metabolism, and improves brain function in some rodent models of neurodegeneration. We conducted a placebo-controlled randomized pilot study with the primary objective of determining safety of NR in older adults with mild cognitive impairment (MCI). Twenty subjects with MCI were randomized to receive placebo or NR using dose escalation to achieve, and maintain, a final dose of 1 g/day over a 10-week study duration. The primary outcome was post-treatment change from baseline measures of cognition (Montreal Cognitive Assessment, MoCA). Predefined secondary outcomes included post-treatment changes in cerebral blood flow (CBF); blood NAD+ levels; and additional neurocognitive, psychometric, and physical performance tests. DNA methylation was assessed in peripheral blood mononuclear cells (PBMCs) as an exploratory outcome. The target NR dose was safely achieved as evidenced by a 2.6-fold increase in blood NAD+ in the NR group (p < 0.001, 95% CI [17.77, 43.49]) with no between-group difference in adverse event reporting. MoCA and other neurocognitive and psychometric metrics remained stable throughout the study. NR reduced CBF in the default mode network (DMN) with greatest differences observed in the left inferior parietal lobe (IPL) (DMN p = 0.013, μ = 0.92, 95% CI [0.23, 1.62]; left IPL p = 0.009, μ = 1.66, 95% CI [0.5, 2.82]). Walking speed in the placebo group significantly improved across the study duration suggestive of a practice effect but did not change in the NR group (p = 0.0402 and p = 0.4698, respectively). Other secondary outcome measures remained stable. Global methylation analyses indicated a modest NR-associated increase in DNA methylation and concomitant reduction in epigenetic age as measured by PhenoAge and GrimAge epigenetic clock analyses. In summary, NR significantly increased blood NAD+ concentrations in older adults with MCI. NR was well tolerated and did not alter cognition. While CBF was reduced by NR treatment, statistical significance would not have withstood multiple comparisons correction. A larger trial of longer duration is needed to determine the potential of NR as a strategy to improve cognition and alter CBF in older adults with MCI. ClinicalTrials.gov NCT02942888
Preprint
Full-text available
Identifying transcriptional enhancers and their target genes is essential for understanding gene regulation and the impact of human genetic variation on disease. Here we create and evaluate a resource of >13 million enhancer-gene regulatory interactions across 352 cell types and tissues, by integrating predictive models, measurements of chromatin state and 3D contacts, and large-scale genetic perturbations generated by the ENCODE Consortium. We first create a systematic benchmarking pipeline to compare predictive models, assembling a dataset of 10,411 element-gene pairs measured in CRISPR perturbation experiments, >30,000 fine-mapped eQTLs, and 569 fine-mapped GWAS variants linked to a likely causal gene. Using this framework, we develop a new predictive model, ENCODE-rE2G, that achieves state-of-the-art performance across multiple prediction tasks, demonstrating a strategy involving iterative perturbations and supervised machine learning to build increasingly accurate predictive models of enhancer regulation. Using the ENCODE-rE2G model, we build an encyclopedia of enhancer-gene regulatory interactions in the human genome, which reveals global properties of enhancer networks, identifies differences in the functions of genes that have more or less complex regulatory landscapes, and improves analyses to link noncoding variants to target genes and cell types for common, complex diseases. By interpreting the model, we find evidence that, beyond enhancer activity and 3D enhancer-promoter contacts, additional features guide enhancer-promoter communication including promoter class and enhancer-enhancer synergy. Altogether, these genome-wide maps of enhancer-gene regulatory interactions, benchmarking software, predictive models, and insights about enhancer function provide a valuable resource for future studies of gene regulation and human genetics.
Article
Full-text available
Functional decline of physiological systems during ageing leads to age-related diseases. Dietary glycine increases healthy lifespan in model organisms and might decrease inflammation in humans, suggesting its geroprotective potential. This review summarises the evidence of glycine administration on the characteristics of eleven physiological systems in adult humans. Databases were searched using key search terms: ‘glycine’, ‘adult’, ‘supplementation’/ ‘administration’/ ‘ingestion’/ ‘treatment’. Glycine was administered to healthy and diseased populations (18 and 34 studies) for up to 14 days and 4 months, respectively. The nervous system demonstrated the most positive effects, including improved psychiatric symptoms from longer-term glycine administration in psychiatric populations. While longer-term glycine administration improved sleep in healthy populations, these studies had small sample sizes with a high risk of bias. Larger and long-term studies with more robust study designs in healthy populations to examine the effects of glycine administration on preventing, delaying or reversing the ageing process are warranted.
Article
Full-text available
Young blood plasma is known to confer beneficial effects on various organs in mice and rats. However, it was not known whetherplasma from young adult pigs rejuvenates old rat tissues at the epigenetic level; whether it alters the epigenetic clock, which is a highlyaccurate molecular biomarker of aging. To address this question, we developed and validated six different epigenetic clocks for rattissues that are based on DNA methylation values derived from n = 613 tissue samples. As indicated by their respective names, the ratpan-tissue clock can be applied to DNA methylation profiles from all rat tissues, while the rat brain, liver, and blood clocks apply to thecorresponding tissue types. We also developed two epigenetic clocks that apply to both human and rat tissues by adding n = 1366human tissue samples to the training data. We employed these six rat clocks to investigate the rejuvenation effects of a porcine plasmafraction treatment in different rat tissues. The treatment more than halved the epigenetic ages of blood, heart, and liver tissue. A lesspronounced, but statistically significant, rejuvenation effect could be observed in the hypothalamus. The treatment was accompanied byprogressive improvement in the function of these organs as ascertained through numerous biochemical/physiological biomarkers,behavioral responses encompassing cognitive functions. An immunoglobulin G (IgG) N-glycosylation pattern shift from pro- to anti-inflammatory also indicated reversal of glycan aging. Overall, this study demonstrates that a young porcine
Article
The vast majority of missense variants observed in the human genome are of unknown clinical significance. We present AlphaMissense, an adaptation of AlphaFold fine-tuned on human and primate variant population frequency databases to predict missense variant pathogenicity. By combining structural context and evolutionary conservation, our model achieves state-of-the-art results across a wide range of genetic and experimental benchmarks, all without explicitly training on such data. The average pathogenicity score of genes is also predictive for their cell, capable of identifying short essential genes that existing statistical approaches are underpowered to detect. As a resource to the community, we provide a database of predictions for all possible human single amino acid substitutions and classify 89% of missense variants as either likely benign or likely pathogenic.