Mol2Context-vec: learning molecular representation from context awareness for drug discovery

Abstract

With the rapid development of proteomics and the rapid increase of target molecules for drug action, computer-aided drug design (CADD) has become a basic task in drug discovery. One of the key challenges in CADD is molecular representation. High-quality molecular expression with chemical intuition helps to promote many boundary problems of drug discovery. At present, molecular representation still faces several urgent problems, such as the polysemy of substructures and unsmooth information flow between atomic groups. In this research, we propose a deep contextualized Bi-LSTM architecture, Mol2Context-vec, which can integrate different levels of internal states to bring dynamic representations of molecular substructures. And the obtained molecular context representation can capture the interactions between any atomic groups, especially a pair of atomic groups that are topologically distant. Experiments show that Mol2Context-vec achieves state-of-the-art performance on multiple benchmark datasets. In addition, the visual interpretation of Mol2Context-vec is very close to the structural properties of chemical molecules as understood by humans. These advantages indicate that Mol2Context-vec can be used as a reliable and effective tool for molecular expression. Availability: The source code is available for download in https://github.com/lol88/Mol2Context-vec.

Multitask deep learning with dynamic task balancing for quantum mechanical properties prediction

“…To reduce the computational cost, several studies adopted multitask learning to predict multiple QMPs simultaneously. 16,29,30 For example, MoleculeNet 16 compared several machine learning methods trained in the multitask setting for QMPs prediction and found that graph-based methods outperform traditional machine learning methods relying on hand-crafted features. Attentive FP 29 introduced a novel GNN that uses graph attention mechanisms at both the atom and molecule levels to learn both local and nonlocal properties of a given chemical structure.…”

“…We compared the proposed method with the existing state-ofthe-art (SOTA) multitask learning models ECFP, 16 CM, 16 GC, 16 DTNN, 50 MPNN, 12 AttentiveFP, 29 and Mol2Context. 30 Note that we excluded methods 14,15 that were implemented in single task settings in this group of experiments. Alternatively, we tailored the SOTA single task learning model TrimNet 15 to a multitask learning model by modifying the output single unit into multiple units.…”

“…We used mean absolute error (MAE) to evaluate the performance of different models following previous studies. [14][15][16]29,30 For simplicity, we denoted MDGNN with DTB as MDGNN (DTB). Tables 3 and 4 summarize the comparison of different multitask learning models in QM9 and Alchemy, respectively.…”

“…[23][24][25][26] On the other hand, attention mechanisms have been widely adopted in the biomedicine field to rationalize the learningbased models. 15,[27][28][29][30] However, the attention-based visual explanation methods are unable to provide interpretability in multitask learning scenarios since multiple output properties correspond to the same attention map.…”

On modeling and utilizing chemical compound information with deep learning technologies: A task-oriented approach

“…Recurrent Neural Network (RNN) is useful for learning relational data or capturing sequential/temporal information because output from the previous state is fed into the current state. Similar to CNNs, RNNs have also been widely used to investigate sequentially formalized molecular representations (e.g., SMILES) [106] , [127] . Compared to CNN (See Section 3.1 ), RNN and its variants (e.g., LSTMs or GRUs) can capture long-range relationships among chemical elements in SMILES due to their innate recurrent memory mechanism.…”