Back

MERIT: Mechanism driven model predicts drug outcomes and nominates indications for failed drugs

Koh-Tan, H. H. C.; Meic, I.; Sarı, B. A.; Muller, S.; Richman, G.

2026-08-14 bioinformatics
10.64898/2026.08.09.743088 bioRxiv
Show abstract

Drug development depends on efficacy and safety, but many trial-outcome prediction models incorporate trial design, prior development history or compound identity, enabling compound memorization and inflating apparent performance. We developed MEchanism-Resolved Inference of Trial outcomes (MERIT), a model that predicts trial outcomes from molecular and disease features without using information on similar-compound success. MERIT integrates the disease and drug of interest with large-scale drug-protein, protein-metabolite and immune interaction maps to link a drugs intended and potential off-target effects to tissue-specific efficacy and safety. Across 753 small-molecule drugs and 3,133 trials, MERIT achieved a best-in-class overall AUROC of 0.770 (0.765 for efficacy and 0.784 for safety). MERIT also recovered the eventual approved indications for 83% of failed drugs. Finally, we registered locked, outcome-blind predictions for 55 drug-indication pairs in ongoing Phase III trials, establishing a prospective evaluation cohort.

Matching journals

The top 8 journals account for 50% of the predicted probability mass.

1
Nature Communications
○ 5641 papers in training set
Top 5%
25.9%
Similar papers in this journal
2
Cell Systems
○ 201 papers in training set
Top 0.8%
5.3%
Similar papers in this journal
3
Science
● 477 papers in training set
Top 2%
4.2%
Similar papers in this journal
4
Nature Machine Intelligence
○ 70 papers in training set
Top 0.7%
3.9%
Similar papers in this journal
5
Science Advances
● 1243 papers in training set
Top 11%
3.2%
Similar papers in this journal
6
Nature
○ 645 papers in training set
Top 5%
3.1%
7
Nature Biomedical Engineering
○ 47 papers in training set
Top 0.4%
3.1%
8
Briefings in Bioinformatics
◐ 354 papers in training set
Top 3%
2.7%
50% of probability mass above
9
PLOS ONE
● 5266 papers in training set
Top 39%
2.7%
10
eLife
● 5828 papers in training set
Top 43%
2.4%
11
Nature Methods
○ 385 papers in training set
Top 4%
2.3%
12
Journal of Chemical Information and Modeling
● 238 papers in training set
Top 1%
2.3%
13
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 22%
2.3%
14
Communications Biology
○ 993 papers in training set
Top 11%
2.1%
15
Molecular Systems Biology
○ 162 papers in training set
Top 1%
1.9%
16
Nature Genetics
○ 286 papers in training set
Top 3%
1.6%
17
Nature Chemical Biology
○ 119 papers in training set
Top 2%
1.6%
18
Nucleic Acids Research
◐ 1281 papers in training set
Top 10%
1.5%
19
Communications Chemistry
○ 48 papers in training set
Top 0.6%
1.5%
20
Scientific Reports
○ 3612 papers in training set
Top 59%
1.4%
21
Bioinformatics
◐ 1204 papers in training set
Top 7%
1.4%
22
PLOS Computational Biology
● 1863 papers in training set
Top 16%
1.3%
23
iScience
○ 1154 papers in training set
Top 24%
1.3%
24
Nature Biotechnology
○ 172 papers in training set
Top 3%
1.1%
25
Cell Genomics
○ 172 papers in training set
Top 4%
1.1%
26
Nature Structural & Molecular Biology
236 papers in training set
Top 3%
1.0%
27
Patterns
○ 78 papers in training set
Top 3%
0.8%
28
Genome Biology
○ 637 papers in training set
Top 9%
0.8%
29
eBioMedicine
○ 183 papers in training set
Top 7%
0.8%
30
Cancer Cell
○ 42 papers in training set
Top 2%
0.6%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.