PGMG: A Pharmacophore-Guided Deep Learning Approach for Bioactive Molecular Generation

Abstract

The rational design of novel molecules with desired bioactivity is a critical but challenging task in drug discovery, especially when treating a novel target family or understudied targets. Here, we propose PGMG, a pharmacophore-guided deep learning approach for bioactivate molecule generation. Through the guidance of pharmacophore, PGMG provides a flexible strategy to generate bioactive molecules matching given pharmacophore models. PGMG uses a graph neural network to encode pharmacophore features and spatial information and a transformer decoder to generate molecules. A latent variable is introduced to solve the many-to-many mapping between pharmacophores and molecules and improve the diversity of generated molecules. In addition, these generated molecules are of high validity, uniqueness, and novelty. In the case studies, we demonstrate using PGMG in ligand-based and structure-based drug de novo design, as well as in lead optimization scenarios. Overall, the flexibility and effectiveness make PGMG a useful tool to accelerate drug discovery process.

CLIP-Gaze: Towards General Gaze Estimation via Visual-Linguistic Model

“…Vision-Language Model Recently, many works have taken advantage of CLIP's flexible text manipulation and visual alignment capabilities to enhance the open detection or generalization performance of specific tasks, such as Det-CLIP (Yao et al 2022), DenseCLIP (Rao et al 2022), CLIP-Gap (Vidit, Engilberge, and Salzmann 2023), Ordinalclip (Li et al 2022), CLIP-Cluster (Shen et al 2023) and so on. Furthermore, to improve the performance of vision-language models on downstream tasks, a more effective approach is to learn continuous text prompts by text prompt tuning (Zhou et al 2022b,a).…”

Landscape quality assessments using deep learning

“…While much attention has been devoted to classification tasks using VLMs and dataefficient learning methods (Radford et al, 2021;Zhou et al, 2022a;Gabeff et al, 2023), comparatively little has been done to develop methods compatible with regression tasks. Li et al introduced OrdinalCLIP, which uses an ordinal output space in order to utilise the classification-based few-shot methods (Li et al, 2022). Hentschel et al (Hentschel et al, 2022) trained a linear probe on CLIP image features on the regression task of image photographic aesthetics understanding in a few-shot setting, with competitive results compared to a fully-trained baseline.…”