PosterPDF Available

A FAST AND NON-INVASIVE WORKFLOW FOR THE DETECTION OF CLINICAL OUTCOMES ON MASS SPECTRAL DATA FROM URINE: APPLICATION TO ANEUPLOIDIES IN HIGH-RISK PREGNANCY

Authors:
  • Bioenhancer Systems & CiiEM (Egas Moniz)
  • SAB Scientific Consultancy Ltd
  • MAP Sciences Ltd, UK

Abstract

QUESTION – Is it possible to developed a fully automated computational tool for the detection of aneuploidies from mass spectral data of high-risk pregnancy from UK? ANSWER – Predictive models generated with a fully automated computational tool were able to detect generic aneuploidies and trisomy 21 in first Trimester maternal pregnancy urine MALDI-TOF mass spectral data.
QUESTION Is it possible to developed a fully automated
computational tool for the detection of aneuploidies from mass
spectral data of high-risk pregnancy from UK?
ANSWER Predictive models generated with a fully automated
computational tool were able to detect generic aneuploidies and
trisomy 21 in first Trimester maternal pregnancy urine MALDI-TOF
mass spectral data.
BACKGROUND MALDI-ToF generated urine mass spectra is anon-
invasive approach shown to have potential for mass market
diagnostic or screening tests, with relevance to obstetrics and
reproductive medicine. However, urine mass spectra of large data
sets are too complex and time consuming to be tackled with
conventional statistical methods. Thus, automated computational-
based workflows are needed. A previous study on a small data set
of urine mass spectra from high-risk pregnant in UK showed
promising results for screening of aneuploidies.
DESIGN - A computational workflow was developed to perform an
automated mass spectra quality assessment, comparative analysis,
and generation of predictive models (Figure 1). Workflow
robustness for pre-processing and quality control was tested
against 10000 mass spectra. The workflow was further applied to
438 mass spectra of high-risk pregnancy urine, previously collected
from King's College Hospital (UK). Generic aneuploidies and trisomy
21 pregnancy-outcomes were the tested groups, whereas non-
aneuploid the control group. In generic aneuploidies trisomies 21,
18,16 and 13 were included.
METHODS Urine was collected prospectively from patients
attending a high risk antenatal care clinic at Kings College London,
identification blinded and analysed at MAP Diagnostics Laboratory
(Bedford, UK) on Shimadzu Axima MALDI MS. The computational
workflow was developed in house using Python 2.7. Predictive
models based on identification of mass spectral patterns were
systematically generated using a subset of the data (training set).
Models were tested retrospectively by accessing the ROC on the
remaining data and computing optimal sensitivities, false positive
rates (1- specificity) and AUC.
RESULTS The computational workflow was able to process 10000
urine samples in a total of 4 hours. The computational workflow was
accurate in quality control decisions on urine spectra, in
agreement with independently made manual inspection of 3000
samples. On a subset of the data with known clinical outcomes,
the computational tool generated and tested a total of 4600
pattern based predictive models for detecting Trisomy 21 and
generic aneuploidies in only 37 min.
Ricardo J. Pais1, Kypros Nicolaides2, Stephen A. Butler1, Raminta Zmuidinaite1, Sholeh
Keshavarz1, Ray Iles1,3,
A FAST AND NON-INVASIVE WORKFLOW FOR THE DETECTION OF
CLINICAL OUTCOMES ON MASS SPECTRAL DATA FROM URINE:
APLICATION TO ANEUPLOIDIES IN HIGH-RISK PREGNANCY.
1. MAP Sciences, London, UK & iLab, Bedford, UK. 2. Harris Birthright Centre, Kings College Hospital London UK. 3.
Dean’s office, College of Healthcare Sciences, Abu Dhabi University, Abu Dhabi UAE
CONCLUSION Fast and fully automated tool for extracting
diagnostic potential from urine without relying on traditional
biomarker. This tool saves time, money and human resources
which enables fast and cheaper screening tests for larger
populations world-wide regardless of economic status.
Several models obtained satisfactory performance with
sensitivities from 75-100% and false positive rates lower than
20% (Figure 2). The best performance obtained for Trisomy
21 detection was 80% sensitivity with 12% false positives,
whereas for generic aneuploidies the sensitivity was 75% with
6% false positives. In general, models performance
depended on the choice of the m/z window for pattern
identification, with an optimal value of 10 m/z (Figure 2).
Supported by
Figure 2 Optimal performance of predictive models with
different m/z window size.
Figure 1 Workflow for systematic predictive model
generation and validation, including sample processing in
the wet laboratory (7 min per sample) and data analysis in a
dry laboratory (42 files/min).
T21 detection Aneuploidies detection
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.