A multimodal representation learning platform for accurate molecular ADMET prediction
Accurate ADMET prediction is essential for prioritizing compounds before costly experimental validation, yet ADMET tasks are highly heterogeneous. Properties such as solubility, permeability, protein binding, clearance, transporter activity and toxicity are governed by different molecular signals, ranging from local functional groups and physicochemical descriptors to bonded topology and…
Accurate prediction of ADMET properties is crucial for selecting compounds for further experimental validation. However, ADMET tasks encompass a wide range of molecular signals, including local functional groups, physicochemical descriptors, bonded topology, and three-dimensional geometry. As a result, a single molecular representation may not be optimal for all ADMET tasks.
The researchers introduce Trimole-Hybrid, a task-wise multimodal framework designed to tackle the heterogeneity of ADMET tasks. Trimole-Hybrid creates a pool of predictors using complementary molecular representations, such as SMILES, graph-based, geometry-sensitive, EPT/3D, and chemical descriptors. For each ADMET task, the framework selects the best-performing predictor to generate the final prediction.
When tested on 22 Therapeutics Data Commons ADMET benchmarks, Trimole-Hybrid outperformed public top-1 methods on 10 tasks and ranked within the top 10 for 21 tasks. In addition, a series of ablation studies demonstrated the importance of both multimodal molecular representations and task-specific ensemble strategies.
Furthermore, the researchers conducted two small-molecule case studies to evaluate Trimole-Hybrid's sensitivity to changes in essential functional motifs. The results indicate that the framework can capture ADMET-relevant molecular substructures, making it a promising tool for accurate molecular ADMET prediction.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
