From Bias to Breakdown: Benchmarking Failure Mode Analysisof Single-cell RNA Sequencing Foundation Models in AcuteMyeloid Leukemia
Bibliographic record
Abstract
Foundation models (FMs) trained on large-scale single-cell RNA-seq (scRNA‐seq) data have shown strong performance across various biological tasks. These performances are often reported across a large set of test benchmarks across all samples. However, the pretraining data of these models are often highly imbalanced across disease types, patients' conditions, and demographics. For instance, disease samples are rarer and more challenging to collect, and the pretraining sets contain many more healthy cells. Such imbalances can hurt performance on underrepresented disease cases and the equality of the model outcome. To evaluate this hypothesis, we benchmark off-the-shelf scRNA-seq foundation models for cell-type classification in acute myeloid leukemia (AML), a rare but clinically important disease that represents low-prevalence settings. Here, besides overall performance, we conduct subgroup analysis of the outcome across cell types and disease conditions (clinical timepoints). Our results suggest that despite high overall F1 scores in cell-type classification, performance drops in disease conditions and varies across cell types. These findings highlight a limitation of current scRNA-seq foundation models and motivate more balanced pretraining and failure mode analysis rather than an overall performance report.
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 | 0.005 | 0.009 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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".