Refining sequence-to-activity models by increasing model resolution
Bibliographic record
Abstract
Decoding the cis-regulatory syntax that controls gene expression is essential for improving our understanding of cell differentiation and disease. To identify regulatory motifs and their regulatory syntax, deep learning based sequence-to-activity (S2A) models learn transcription factor binding motifs and their combinations from DNA sequence by modeling measured chromatin accessibility. Previously, we developed AI-TAC, a S2A model that predicts chromatin accessibility across various immune cell types in multi-task fashion, effectively decoding the regulatory syntax underlying immune cell differentiation. While ATAC-seq is commonly used to measure regional accessibility, it also provides high-resolution profiles, the distribution of Tn5 insertion sites, that offer additional insights into the precise location and strength of TF binding sites. Here we demonstrate that modeling ATAC-seq profiles alongside accessibility consistently improves predictions of differential chromatin accessibility across cell types. Moreover, we also find that multi-task learning across related immune cell types consistently outperforms single-task models. To understand what additional information bpAITAC learns from ATAC-seq profiles, we systematically compare sequence attributions from models trained with and without ATAC-seq profiles. We identify novel motifs with strong effect sizes that emerge only when profile data is included. Our findings suggest that modeling ATAC-seq at base-pair resolution enables the model to learn a more nuanced and sensitive representation of the cis-regulatory syntax driving immune cell-specific chromatin landscapes.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 teacher head, 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".