Learning a CoNCISE language for small-molecule binding
Bibliographic record
Abstract
Abstract Rapid advances in deep learning have improved in silico methods for drug-target interaction (DTI) prediction. However, current methods do not scale to the massive catalogs that list millions or billions of commercially-available small molecules. Here, we introduce CoNCISE, a method that accelerates drug-target interaction (DTI) prediction by 2-3 orders of magnitude while maintaining high accuracy. CoNCISE uses a novel vector-quantized codebook approach and a residual-learning based training of hierarchical codes. Strikingly, we find that much of binding-specificity information in the small molecule space can be compressed into just 15 bits of information per compound, characterizing all small molecules into 32,768 hierarchically-organized binding categories. Our DTI architecture, which combines these compact ligand representations with fixed-length protein embeddings in a cross-attention framework, achieves state-of-the-art prediction accuracy at unprecedented speed. We demonstrate CoNCISE’s practical utility by indexing 6.4 billion ligands in the Enamine dataset, enabling researchers to query vast chemical libraries against a protein target in seconds. A “CoNCISE + docking” pipeline screened Enamine to propose strong binders (predicted K D ≈ 10-20 µ M) of three difficult-to-drug targets, each within two hours. CoNCISE’s advance could democratize access to largescale computational drug discovery, potentially enabling rapid identification of promising molecules for therapeutic targets and cellular perturbations.
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 | 0.001 | 0.006 |
| 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 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.007 | 0.003 |
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".