Adversarial training improves model interpretability in single-cell RNA-seq analysis
Bibliographic record
Abstract
Motivation: Predictive computational models must be accurate, robust, and interpretable to be considered reliable in important areas such as biology and medicine. A sufficiently robust model should not have its output affected significantly by a slight change in the input. Also, these models should be able to explain how a decision is made to support user trust in the results. Efforts have been made to improve the robustness and interpretability of predictive computational models independently; however, the interaction of robustness and interpretability is poorly understood. Results: As an example task, we explore the computational prediction of cell type based on single-cell RNA-seq data and show that it can be made more robust by adversarially training a deep learning model. Surprisingly, we find this also leads to improved model interpretability, as measured by identifying genes important for classification using a range of standard interpretability methods. Our results suggest that adversarial training may be generally useful to improve deep learning robustness and interpretability and that it should be evaluated on a range of tasks. Availability and implementation: Our Python implementation of all analysis in this publication can be found at: https://github.com/MehrshadSD/robustness-interpretability. The analysis was conducted using numPy 0.2.5, pandas 2.0.3, scanpy 1.9.3, tensorflow 2.10.0, matplotlib 3.7.1, seaborn 0.12.2, sklearn 1.1.1, shap 0.42.0, lime 0.2.0.1, matplotlib_venn 0.11.9.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.008 | 0.026 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".