Back

Widespread data leakage inflates performance estimates in cancer drug response prediction

Asiaee, A.; Strauch, J.; Azinfar, L.; Pal, S.; Pua, H. H.; Long, J. P.; Coombes, K. R.

2026-02-08 bioinformatics
10.64898/2026.02.05.704016 bioRxiv
Show abstract

Drug response prediction models are widely used to nominate biomarkers and guide preclinical drug prioritization. However, their reported performance hinges on rigorous separation of training and test data during cross-validation (CV). Here we show that a commonly used pattern--supervised feature screening performed on the full dataset before CV--introduces data leakage that systematically underestimates prediction error. Analyzing 265 drugs across 1,462 cancer cell lines, we find that leakage-free CV increases mean squared error (MSE) by 16.6%, with low feature-set overlap between leaked and leakage-free pipelines (mean Jaccard 0.18). A manual audit of 12 recent deep learning and classical methods found confirmed leakage in 10. Such inflated performance estimates likely contribute to computational predictions that fail during independent validation or experimental follow-up. We provide an audit guide and reference implementation to prevent leakage, and introduce a tissue-aware Data Shared Elastic Net (DSEN) that, under correct evaluation, improves prediction for 65.7% of drugs while yielding sparser, more targeted biomarker sets.

Matching journals

The top 10 journals account for 50% of the predicted probability mass.

1
Nature Machine Intelligence
61 papers in training set
Top 0.1%
10.6%
2
Cell Systems
167 papers in training set
Top 1.0%
10.3%
3
Nature Communications
4913 papers in training set
Top 22%
8.6%
4
Scientific Reports
3102 papers in training set
Top 30%
4.0%
5
Nature Biomedical Engineering
42 papers in training set
Top 0.2%
4.0%
6
Bioinformatics
1061 papers in training set
Top 5%
3.6%
7
Advanced Science
249 papers in training set
Top 7%
2.9%
8
Proceedings of the National Academy of Sciences
2130 papers in training set
Top 24%
2.9%
9
Communications Biology
886 papers in training set
Top 4%
2.6%
10
Nature Methods
336 papers in training set
Top 3%
2.6%
50% of probability mass above
11
Patterns
70 papers in training set
Top 0.4%
2.4%
12
npj Digital Medicine
97 papers in training set
Top 2%
2.1%
13
Briefings in Bioinformatics
326 papers in training set
Top 3%
2.1%
14
PLOS ONE
4510 papers in training set
Top 50%
1.9%
15
Genome Medicine
154 papers in training set
Top 4%
1.8%
16
Cell Reports Medicine
140 papers in training set
Top 3%
1.7%
17
Nature Medicine
117 papers in training set
Top 2%
1.7%
18
PLOS Computational Biology
1633 papers in training set
Top 16%
1.7%
19
Cell Genomics
162 papers in training set
Top 3%
1.7%
20
Nature Biotechnology
147 papers in training set
Top 5%
1.4%
21
Journal of Chemical Information and Modeling
207 papers in training set
Top 2%
1.4%
22
Science Advances
1098 papers in training set
Top 21%
1.4%
23
npj Systems Biology and Applications
99 papers in training set
Top 1%
1.2%
24
iScience
1063 papers in training set
Top 21%
1.2%
25
Cancer Cell
38 papers in training set
Top 1%
0.9%
26
Artificial Intelligence in the Life Sciences
11 papers in training set
Top 0.2%
0.9%
27
GigaScience
172 papers in training set
Top 2%
0.9%
28
BMC Bioinformatics
383 papers in training set
Top 6%
0.9%
29
eLife
5422 papers in training set
Top 55%
0.8%
30
npj Precision Oncology
48 papers in training set
Top 1%
0.8%