Pre-submission inquiry for the publication of DeCovarT

[bastienchassagnol]

Author: Bastien Chassagnol (@bastienchassagnol)
ORCID: 0000-0002-8955-2391
Repository: https://github.com/bastienchassagnol/DeCovarT
Version: 2.0.1
Documentation: https://bastienchassagnol.github.io/DeCovarT/
Method preprint: https://arxiv.org/abs/2309.09557

---

Summary

DeCovarT estimates cell-type proportions in bulk RNA-seq samples using a
probabilistic convolution model that incorporates the gene–gene covariance
structure of cellular profiles, inferring using high-dimensional, sparsity-aware inference techniques.
Ratios are recovered by maximum likelihood under a multivariate Gaussian convolution, with analytic gradients
and Hessians, additive log-ratio reparametrisation on the simplex, and
Marquardt–Levenberg or Newton-type optimisers.

The package also ships classical deconvolution baselines (OLS, NNLS,
DeconRNASeq-style QP, robust linear regression, CIBERSORT-style ν-SVR) for benchmarking,
and evaluation utilities.

---

Scope question

I believe DeCovarT falls under statistical software rather than the
general categories in the policies.
Could the editors confirm whether it is in scope for rOpenSci software peer
review, and whether the statistical-software standards (including srr) apply?

Target audience: computational biologists and immunologists working with
bulk RNA-seq who need reproducible cell-type deconvolution with uncertainty
awareness via the covariance structure.

Scientific applications: cell composition in bulk transcriptomes,
and more realistic synthetic validation scenarios, modelling interactions between genes
through a variety of random networks generators.

---

Overlap with existing software

Several R/Bioconductor packages address bulk deconvolution, including
reference-based methods such as MuSiC, BayesPrism, BisqueRNA,
SCDC, InstaPrism, and signature-based tools (e.g. EPIC,
CIBERSORT implementations). DeCovarT differs by:

1. Modelling: multivariate Gaussian convolution MLE with explicit
   covariance structure, while most standard approaches assume independence
   between genes.
2. Optimisation: constrained MLE with analytic derivatives and simplex
   reparametrisation, enabling the derivation of asymptotic intervals using an explicit form of the expected Fisher information matrix.

I am happy to expand the README “similar packages” section if editors consider
that helpful before a full submission.

---

Publication and review timing

I plan to submit a methods article to Nature Methods describing the
methodology and validation. That manuscript concerns the scientific method,
not a separate software paper about the R package itself.

• The package is not currently under review at another software-review
  venue.
• The Nature Methods manuscript is not yet submitted;

Please advise whether this sequencing is acceptable.

---

Package readiness (pkgcheck)

I have run pkgcheck::pkgcheck() locally (with GitHub API access). All
mandatory checks pass, including CI
(latest run),
77.8% test coverage, and clean R CMD check.

Remaining optional :eyes: items: a relatively large number of Imports (20);
I can address these during review if the package
is in scope, most of these imports being related to using functionalities of the tidyverse ecosystem,
instead of R base.

Lifecycle: Active / stable
License: MIT
Maintainer commitment: I expect to maintain the package for ≥2 years.

---

Questions for the editors

1. Is DeCovarT in scope as statistical software?
2. Given overlap with existing deconvolution packages, is the covariance-aware
   MLE framing sufficient differentiation for the rOpenSci suite?
3. Is the planned Nature Methods article (method paper, not a JOSS-style
   software note) compatible with rOpenSci review if timed as above?

Thank you for your guidance.

[mpadge]

Thansk @bastienchassagnol for your inquiry, and particularly for the detailed information you've provided, and comparisons to similar software. I'll respond to your questions below, but first could you please edit your initial comment used to open the issue and replace with our template?

It's probably easiest to:

1. Copy that template somewhere locally;
2. Copy what you've currently written across to the appropriate locations
3. Edit opening comment to delete current structure and paste in updated structure that matches template form?

Please also respond to the final checkbox about possible use of generative AI tools.

---

Responses to your questions

1. Is DeCovarT in scope as statistical software?

Scope decisions for statistical software are largely based on whether software is able to comply with at least 50% of general standards plus at least 50% for the chosen category/categories. You'll need to estimate that yourself, and ultimately insert the compliance statements for your chosen category or categories before we proceed to a full review.

2. Given overlap with existing deconvolution packages, is the covariance-aware MLE framing sufficient differentiation for the rOpenSci suite?

As long as you can comply with at least half of all standards, it should be fine. Explicit control over covariance structures is something we consider important enough that that alone should suffice to consider your package as making a unique contribution.

3. Is the planned Nature Methods article (method paper, not a JOSS-style
   software note) compatible with rOpenSci review if timed as above?

We have no strict requirements, but we do strongly recommending completing our review process prior to submitting to any subsequent venues. This recommendation is based on our experiences that the review process frequently results in significant changes to software. While your purported paper sounds quite distinct from the package that would be reviewed here, reviews often consider the meta-structure of workflows which packages are designed to support, and that could ultimately impact any paper which described your methodology.

---

Summary

In short: Your package seems like it could be in shape, but you'll need to:

1. Edit the initial issue comment to conform with our expected template structure; and
2. Read through the standards for any category or categories you think best fit your package, and estimate whether you think you'll be able to comply with at least half of all standards.

Once you've done that, and we have an idea of which statistial category your package fit within, we'll happily continue the discussion here.

Thanks again for considering submitting your package here!