Search

Self-supervised pre-trained language models have recently risen as a powerful approach in learning protein representations, showing exceptional effectiveness in various biological tasks, such as drug discovery. Amidst the evolving trend in protein language model development, there is an observable shift towards employing large-scale multimodal and multitask models. However, the predominant reliance on empirical assessments using specific benchmark datasets for evaluating these models raises concerns about the comprehensiveness and efficiency of current evaluation methods. Addressing this gap, our study introduces a novel quantitative approach for estimating the performance of transferring multi-task pre-trained protein representations to downstream tasks. This transferability-based method is designed to quantify the similarities in latent space distributions between pre-trained features and those fine-tuned for downstream tasks. It encompasses a broad spectrum, covering multiple domains and a variety of heterogeneous tasks. To validate this method, we constructed a diverse set of protein-specific pre-training tasks. The resulting protein representations were then evaluated across several downstream biological tasks. Our experimental results demonstrate a robust correlation between the transferability scores obtained using our method and the actual transfer performance observed. This significant correlation highlights the potential of our method as a more comprehensive and efficient tool for evaluating protein representation learning.

The development of new drugs is crucial for protecting humans from disease. In the past several decades, target-based screening has been one of the most popular methods for developing new drugs. This method efficiently screens potential inhibitors of a target protein in vitro, but it frequently fails in vivo due to insufficient activity of the selected drugs. There is a need for accurate computational methods to bridge this gap. Here, we present a novel graph multi-task deep learning model to identify compounds with both target inhibitory and cell active (MATIC) properties. On a carefully curated SARS-CoV-2 data set, the proposed MATIC model shows advantages compared with the traditional method in screening effective compounds in vivo. Following this, we investigated the interpretability of the model and discovered that the learned features for target inhibition (in vitro) or cell active (in vivo) tasks are different with molecular property correlations and atom functional attention. Based on these findings, we utilized a Monte Carlo-based reinforcement learning generative model to generate novel multiproperty compounds with both in vitro and in vivo efficacy, thus bridging the gap between target-based and cell-based drug discovery. The tool is freely accessible at https://github.com/SIAT-code/MATIC.