MIRAGE-DTA: conflict-aware multimodal evidence arbitration for drug–target affinity prediction

Clicks: 23
ID: 339509
2026
Article Quality & Performance Metrics
Overall Quality
Not rated
Combines reader engagement with the AI quality analysis. This article has not been analysed, so there is no overall score — reader engagement is measured and shown alongside.
AI Quality Assessment
Not analyzed
Readership in this journal
Emerging

Ranked #951 of 1,006 articles by views in BMC Bioinformatics

Most read Least read

Bar heights use a square-root scale. Only the 120 most-read articles are drawn; the journal has 1,006 in total.

Mint this article as an NFT
Not yet minted

Create a permanent, verifiable on-chain record of this article on the Scimatic Network. The NFT is held in your Journament account, and you can withdraw it to your own wallet at any time.

5 SUSD one-off · no wallet required
Abstract
MOTIVATION: Multimodal drug-target affinity (DTA) models can incorporate molecular, protein, assay and textual evidence whose availability and reliability vary under missing modalities, cold-start entities and assay distribution shift. RESULTS: We present MIRAGE-DTA, a reliability-oriented evidence-arbitration layer that complements DTA encoders by keeping current-sample and training-set retrieval evidence as two explicit chains, carrying modality masks into fusion, retrieving training-only neighbors, and resolving disagreement with an interaction gate and validation-selected anchor. Across BindingDB_Kd, DAVIS, KIBA and a 2010-2024 CHEMBL_ASSAY benchmark with 35,238 interactions, experiments tested cold-start, missing-modality and assay/temporal shift settings. Under 35% sequence/text missingness on CHEMBL_ASSAY, 10-seed paired evaluation showed that MIRAGE-DTA reduced expected calibration error relative to late fusion by 0.0524, 0.0923 and 0.1018 on assay-cold, target-cold and temporal splits. Risk at 80% coverage was significantly lower on target-cold and temporal splits and numerically lower on assay-cold. A complementary complete-evidence direct-pChEMBL regression analysis also showed lower RMSE for the two-branch model than for current-only and retrieval-only regressors across assay-cold, target-cold and temporal splits. Ranking remained broadly competitive, with a modest trade-off on the temporal split, indicating that the principal benefit lies in calibrated and selective prediction under incomplete or shifted evidence. Shortcut audits quantified neighbor-overlap effects and showed that temporal retrieval was not driven by repeated compounds or near-duplicate scaffolds. AVAILABILITY AND IMPLEMENTATION: Source code is available at https://github.com/wushen1103-collab/mirage_drug_miniprototype and archived on Zenodo at https://doi.org/10.5281/zenodo.22695984; data come from public sources.
Reference Key
openalex_W7225725804 Use this key to autocite in the manuscript while using SciMatic Manuscript Manager or Thesis Manager
Authors
Journal BMC Bioinformatics
Year 2026
DOI
10.1093/bioinformatics/btag812
URL
Keywords Keywords not found

Citations

No citations found. To add a citation, contact the admin at info@scimatic.org

No comments yet. Be the first to comment on this article.