Improving drug repositioning with negative data labeling using large language models
Bibliographic record
Abstract
INTRODUCTION: Drug repositioning offers numerous advantages, such as faster development timelines, reduced costs, and lower failure rates in drug development. Supervised machine learning is commonly used to score drug candidates but is hindered by the lack of reliable negative data-drugs that fail due to inefficacy or toxicity- which is difficult to access, lowering their prediction accuracy and generalization. Positive-Unlabeled (PU) learning has been used to overcome this issue by either randomly sampling unlabeled drugs or identifying probable negatives but still suffers from misclassification or oversimplified decision boundaries. RESULTS: We proposed a novel strategy using Large Language Models (GPT-4) to analyze all clinical trials on prostate cancer and systematically identify true negatives. This approach showed remarkable improvement in predictive accuracy on independent test sets with a Matthews Correlation Coefficient of 0.76 (± 0.33) compared to 0.55 (± 0.15) and 0.48 (± 0.18) for two commonly used PU learning approaches. Using our labeling strategy, we created a training set of 26 positive and 54 experimentally validated negative drugs. We then applied a machine learning ensemble to this new dataset to assess the repurposing potential of the remaining 11,043 drugs in the DrugBank database. This analysis identified 980 potential candidates for prostate cancer. A detailed review of the top 30 revealed 9 promising drugs targeting various mechanisms such as genomic instability, p53 regulation, or TMPRSS2-ERG fusion. CONCLUSION: By expanding our negative data labeling approach to all diseases within the ClinicalTrials.gov database, our method could greatly advance supervised drug repositioning, offering a more accurate and data-driven path for discovering new treatments.
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 | 0.004 | 0.008 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| 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".