Maddah, M. | Search bioRxiv

bioRxiv2025

The Latentverse: An Open-Source Benchmarking Toolkit for Evaluating Latent Representations

Self-supervised representation learning is a powerful approach for extracting meaningful features without relying on large amounts of labeled data, making it particularly valuable in fields like healthcare. This enables pretrained models to be shared and fine-tuned with minimal data for various downstream applications. However, evaluating the quality and behavior of these representations remains challenging. To address this, we introduce Latentverse, an open-source library and web-based platform for evaluating latent representations. Latentverse generates detailed reports with visualizations and metrics that provide a comprehensive perspective on different properties of representations, such as clustering, disentanglement, generalization, expressiveness, and robustness. It also allows for the comparison of different representations, enabling developers to refine model architectures and helping users assess how well an embedding model aligns with the requirements of their specific applications. Data and Code AvailabilityThe Latentverse code is available at: https://github.com/broadinstitute/ml4h-latentverse. Institutional Review Board (IRB)This work doesnt require IRB approval.

bioengineering↗

bioRxiv2021

Deep Learning Predicts Patterns of Cardiotoxicity in a High-Content Screen Using Induced Pluripotent Stem Cell-Derived Cardiomyocytes

Drug-induced cardiotoxicity and hepatotoxicity are major causes of drug attrition. To decrease late-stage drug attrition, pharmaceutical and biotechnology industries need to establish biologically relevant models that use phenotypic screening to predict drug-induced toxicity. In this study, we sought to rapidly detect patterns of cardiotoxicity using high-content image analysis with deep learning and induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs). We screened a library of 1280 bioactive compounds and identified those predicted to have cardiotoxic liabilities using a single-parameter score based on deep learning. Compounds with major predicted cardiotoxicity included DNA intercalators, ion channel blockers, epidermal growth factor receptor, cyclin-dependent kinase, and multi-kinase inhibitors. We also screened a diverse library of molecules with unknown targets and identified chemical frameworks with predicted cardiotoxic liabilities. By using this screening approach during target discovery and lead optimization, we can de-risk early-stage drug discovery. We show that the broad applicability of combining deep learning with iPSC technology is an effective way to interrogate cellular phenotypes and identify drugs that protect against diseased phenotypes and deleterious mutations. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=161 HEIGHT=200 SRC="FIGDIR/small/436666v1_ufig1.gif" ALT="Figure 1"> View larger version (42K): org.highwire.dtl.DTLVardef@695707org.highwire.dtl.DTLVardef@1d0980forg.highwire.dtl.DTLVardef@1af30e7org.highwire.dtl.DTLVardef@b3a312_HPS_FORMAT_FIGEXP M_FIG C_FIG CONTRIBUTION TO THE FIELDIn this article, Grafton and colleagues use induced pluripotent stem cell technology and deep learning to train a neural network capable of detecting patterns of cardiotoxicity. To identify bioactive and chemical classes that lead to cardiotoxicity, they combine the neural network with high-content screening of 2560 compounds. The methods described in this study can be used to de-risk early-stage drug development, triage hits, and identify drugs that protect against disease. This screening paradigm will serve as a useful resource for drug discovery and phenotypic interrogation of stem cells and stem cell-derived cell types.

cell biology↗