Molecule-Morphology Contrastive Pretraining for Transferable Molecular Representation

Abstract

Image-based profiling techniques have become increasingly popular over the past decade for their applications in target identification, mechanism-of-action inference, and assay development. These techniques have generated large datasets of cellular morphologies, which are typically used to investigate the effects of small molecule perturbagens. In this work, we extend the impact of such dataset to improving quantitative structure-activity relationship (QSAR) models by introducing Molecule-Morphology Contrastive Pretraining (MoCoP), a framework for learning multi-modal representation of molecular graphs and cellular morphologies. We scale MoCoP to approximately 100K molecules and 600K morphological profiles using data from the JUMP-CP Consortium and show that MoCoP consistently improves performances of graph neural networks (GNNs) on molecular property prediction tasks in ChEMBL20 across all dataset sizes. The pretrained GNNs are also evaluated on internal GSK pharmacokinetic data and show an average improvement of 2.6% and 6.3% in AUPRC for full and low data regimes, respectively. Our findings suggest that integrating cellular morphologies with molecular graphs using MoCoP can significantly improve the performance of QSAR models, ultimately expanding the deep learning toolbox available for QSAR applications.

Unleashing the potential of cell painting assays for compound activities and hazards prediction

“…This limitation can be overcome by using unlabelled compound-image profile data pairs for contrastive learning. Nguyen et al ( Nguyen et al, 2023 ) created a multi-modal contrastive learning framework called Molecule-Morphology Contrastive Pretraining (MoCoP), that integrates molecular graph data and cellular morphology. The authors featurized CP image data from the JUMP dataset with CellProfiler ( Chandrasekaran et al, 2023 ) and used that data to contrastively pre-train a GNN for molecular property prediction.…”

“…The authors featurized CP image data from the JUMP dataset with CellProfiler ( Chandrasekaran et al, 2023 ) and used that data to contrastively pre-train a GNN for molecular property prediction. To evaluate the pre-training, Nguyen et al ( Nguyen et al, 2023 ) measured the accuracy of molecule and image profile retrieval tasks using the JUMP dataset. The retrieval performance was quantified by reporting the average top- k accuracy for retrieving a molecule given its morphology and vice versa .…”

“…However, instead of using CellProfiler features to represent the cell profile images, Sanchez-Fernandez et al ( Sanchez-Fernandez et al, 2023 ) operate on images directly during model training. To evaluate the information content of the embeddings, the authors tried to retrieve images with compounds and vice versa , just like Nguyen et al ( Nguyen et al, 2023 ). On a hold-out dataset of 2115 compound-image pairs, CLOOME was able to correctly retrieve the compound from the image in 3.8% of the cases, and retrieve the image from the compound in 3.2% of the cases.…”