PICASA: Decomposing patient heterogeneity of single-cell cancer data by cross-attention neural networks
Bibliographic record
Abstract
Abstract Motivation Gene expression variation in cancer cells is attributed to many inherited and environmental factors, including genetic variants and cellular landscapes. Decomposing different sources of information is intractable with single-cell RNA-seq alone. However, we show that our new approach, PICASA, can split them with the help of multiple patients, assuming that cell types are widely shared and genetic effects are specifically present in a particular patient. Our approach, based on a cross-attention neural network, was applied to diverse cancer types to identify cell types and patient-specific genetic effects in transcriptomic data. The method highlights residual expressions, excluding cell types, which can implicate patient-specific disease mechanisms. Results PICASA effectively decomposes cell type commonality and the residual patient-specific signature in three different complex cancer datasets, including breast, ovarian, and lung cancers. Unlike many existing distance-based batch adjustment methods (unable to recover cell-type-specific generative models), PICASA learns transferrable gene/feature embedding coordinates and cell-type-specific gene-gene interaction patterns as an attention layer. We also demonstrate that many cancer patient-specific signatures captured by PICASA are deemed somatic and genetic, such as copy number variants (CNV). Availability and implementation PICASA source code is available at https://github.com/causalpathlab/picasa . Additionally, to ensure reproducibility, data analysis and figure generation code is available at https://github.com/causalpathlab/picasa/tree/main/picasa_reproducibility .
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 | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 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".