Content uploaded by Hiroshi Yamada
Author content
All content in this area was uploaded by Hiroshi Yamada on Nov 23, 2022
Content may be subject to copyright.
1
Point-by-point reply to the reviewer’s comments
We greatly appreciate the further comments from the second reviewer. As the reviewer suggests,
we improved our manuscript corresponding to all comments from the reviewer. Below, we have
provided our point-by-point responses to reviewers’ comments.
Manuscript NCOMMS-22-05245A
"A neuronal prospect theory model in the brain reward circuitry"
---------------------------
Reviewers' comments:
---------------------------
Reviewer #1 (Remarks to the Author):
I have no further comments. The authors have satisfactorily addressed my concerns.
Response: We gratefully appreciate the reviewer’s many efforts and useful
comments through this review process.
Reviewer #2 (Remarks to the Author):
Thank you for your revision, which is perfect except for 2 points:
Response: We appreciate the two comments from the reviewer that further
improve our manuscript. We improved our manuscript corresponding to these
comments.
1) In line 196 (all line numbers refer to the revised paper), the authors say ‘percent of variance
explained’ (percent is uncommon to use) but they state a number below 1.0 (which is far too low
for a percentage; this is probably the R2 for the overall regression, or partial R2 for single
regressors in multiple regressions, and an R2 is indeed <1): so maybe say ‘proportion of
variance explained’ or just ‘R2 for variance explained’. The issue comes back in lines 199, 201,
203, 251, fig 4b with line 882/3, fig 5b with line 896.
2
Response: As the reviewer suggest, we replaced “percent variance explained” to
“proportion of variance explained” in the main text and in Figures 4 and 5.
2) Interpretation of the empirical data with the model: In lines 264-275, the authors state that in
all brain structures studied, nonlinear utility and nonlinear probability weighting fitted the
neuronal data better than the EV model (linear utility, linear probability weighting), except for
OFC (line 274). This is an empirical result and thus not an issue. The interpretation at the end of
that Results section is fine as stated: ‘coding underlies subjective valuation under risk’ (line 302).
But I find the interpretation of the model results imprecise and potentially misleading as it
stands: ‘explain the internal subjective valuations of monkeys' (line 330-331)’.
To demonstrate that the model shows subjective as opposed to objective neuronal value
coding, I had suggested in the first round to test whether an EV model (linear utility, linear
probability weighting) would reconstruct the neuronal data less accurately than a Prospect
Theory model (nonlinear utility, nonlinear probability weight). The authors did such an analysis
that resulted in a figure that they decided not to include in the paper (with 3 components in blue,
orange and yellow). I don’t understand that analysis, nor do the authors want to include it in the
paper. Then they ‘calculated the AIC difference between EV model and the objective reward
magnitude and probability model’: the EV model is about objective reward magnitude and
probability, so I don’t really understand what difference they calculated.
So, altogether, the authors do not present better model performance for subjective value
(nonlinear utility, nonlinear probability weight) as compared to objective value (linear utility,
linear probability weighting). Therefore, the model result as it stands does not disambiguate
between subjective coding and objective coding. To avoid overinterpetation, the conclusion
from the model of 'explain the internal subjective valuations of monkeys' (line 330-331) needs
to be amended / qualified / modified here, as well as in the Abstract (line 51), Introduction (line
112) and Discussion (line 354) by saying, for example, that the model also explains objective
value (linear utility, linear probability weighting). This change does not affect the empirical data
and does not change the overall conclusion of the paper.
Response: As the reviewer suggests, we modified our description to tone down at
all the sentences the reviewer pointed out in the results (line 330-331) as well as
the Abstract (line 51), Introduction (line 112) and Discussion (line 354). We
described these changes at the last of this reply. Before explaining our corrections,
we would like to explain our thought for this comment.
3
The main point the reviewer raised in this revision is that overinterpretation would
occur because we did NOT show better model performance for subjective value
(nonlinear utility, nonlinear probability weight) as compared to objective value
(linear utility, linear probability weighting). He said above that “So, altogether, the
authors do not present better model performance for subjective value (nonlinear utility,
nonlinear probability weight) as compared to objective value (linear utility, linear
probability weighting).” We agree to the reviewer’s point that without better model
performance compared to the objective value coding (i.e., EV model), we cannot
say that we found subjective value model coding (i.e., PT models). However, we
have already showed the better model performance for subjective value coding (PT
models) compared to objective value coding (EV model).
In this point of the comment, we think that some miscommunications between
the reviewer and us may occur or we may not precisely follow his/her thought. This
might be because we included other response when we reply to this comment. We
explains these details in the next sections.
We have two objective value coding models. One is the model, which is
represented as the “multiplicative combination” of linear utility and linear probability
weighting. This is the model we have presented originally. The other is the model,
which is represented as the “linear utility and linear probability weighting”, the
reviewer described as above. We previously showed that both models show the
lowest model performances among all subjective and objective coding models in
the previous revision. We will explain again for each of the model performance.
In the first submitted manuscript, we have already showed better model
performance for subjective value coding (PT models) as compared to objective
value coding (EV model, which is multiplicative combination of linear utility and
linear probability weighting), in Figure 4b for a single neuron and Figure 4e for
population of neurons in each of four brain regions. The figure 4e showed the better
model performance for subjective value model compared to the EV model defined
as the multiplicative combination of linear utility and linear probability weighting.
When we performed the point-by-point reply in the first revision, we performed
the analysis modeled by the “linear utility and linear probability weighting” raised by
the reviewer (not the multiplicative EV model that we used in our original analysis).
We compared the model performance for “linear utility and linear probability
weighting” model relative to the multiplicative EV model, same as our analysis in
our manuscript. We obtained the result that “linear utility and linear probability
weighting” did not show better performance compared to the multiplicative EV
4
model. Thus, both multiplicative EV model and “linear utility and linear probability
weighting” model showed the lowest performance to explain the neural activity,
while the subjective value model (PT models) showed better performance.
In this point of the reviewer’s comment, we think that we may not precisely
understand the point of the reviewer related to this model performance. Therefore,
we amended our description to tone down at all the sentences the reviewer pointed
out in the results (line 330-331), as well as in the Abstract (line 51), Introduction
(line 112) and Discussion (line 354) as follows.
Abstract (line 51)
From
“A network model aggregating these signals reliably reconstructed the risk
preferences and subjective probability weighting revealed by the animals’
choices.”
To
“A network model aggregating these signals reconstructed the risk preferences
and subjective probability weighting revealed by the animals’ choices.”
Introduction (line 112)
From
“A simple network model that aggregates these subjective valuation signals,
which are distributed through most parts of the reward circuitry, successfully
reconstructed the monkey’s risk preference and subjective probability
weighting estimated from the choices monkeys made in other situations.”
To
“A simple network model that aggregates these subjective valuation signals,
which are distributed through most parts of the reward circuitry, reconstructed
the monkey’s risk preference and subjective probability weighting estimated
from the choices monkeys made in other situations.”
Results (line 330-331)
From
“Thus, we concluded that a distributed neural code that accumulates individual
5
neuronal signals can explain the internal subjective valuations of monkeys.”
To
“Thus, we concluded that a distributed neural code that accumulates individual
neuronal signals explains the internal subjective valuations of monkeys.”
Discussion (line 354)
From
“suggesting that these subjective valuation signals in the reward circuitry were
integrated into the brain to construct a decision output from risky perspectives.”
To
“suggesting that these subjective valuation signals in the reward circuitry would
be integrated into the brain to construct a decision output from risky
perspectives.”
We appreciate abundant efforts from the reviewer that makes our manuscript
more precise and convincing.