Comprehensive Evaluation of Protein Language Model Embeddings for Drug-Target Affinity Prediction
Accurate identification of drug-target interactions is consequential for novel drug discovery and development. Deep learning methods for drug-target affinity (DTA) prediction have shown great promise in accelerating drug discovery and reducing development costs. Although graph neural networks have improved drug representation learning for DTA prediction tasks, many models still struggle to…
The identification of drug-target interactions is crucial for drug discovery and development. Deep learning methods have proven effective in predicting drug-target affinity (DTA), but many models still struggle with capturing protein information, which hinders prediction accuracy. This study systematically assesses the impact of pre-trained protein language models (PLMs) on DTA prediction tasks.
Four distinct molecular representation backbones were used in multiple experiments, comparing the performance of PLM embeddings against classical 1D convolution methods. Four families of PLMs, each built on unique architectures optimized for various tasks, were integrated into PLM-GraphDTA. Additionally, DeepGraphDTA, an architectural modification of the baseline convolution method, was evaluated.
Both the Davis and KIBA benchmark datasets were used, with concordance index (CI) and mean squared error (MSE) as performance metrics. The models were also tested using cold-start train and test splits, and the contribution of each protein to the total CI was analyzed. The findings suggest that simple architectural modifications to traditional convolution methods may be sufficient to achieve accuracy comparable to large pre-trained PLMs.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
