Abstract 5473: A deep learning framework for extracting comprehensive single-cell biophysical profiles from brightfield images.

Abstract Single-cell biophysical measurements including mass, volume, density, and morphology provide a highly integrative readout of cellular state. We have previously demonstrated that these measurements enable rapid assessment of tumor cell drug response and immune cell functional fitness. As part of a CLIA-certified workflow with a two-day reporting turnaround, this approach enables clinically actionable decision support across diverse solid tumor malignancies and prediction of immune checkpoint blockade response. However, the complex instrumentation and expert operation required for these multiparametric measurements confine their use to CLIA laboratories as LDTs, limiting accessibility in community hospitals and deployment to large diagnostic networks. Here we describe a deep learning approach that estimates single-cell biophysical properties from brightfield imaging data alone. Specifically, we trained a Vector Quantized Variational Autoencoder (VQ-VAE) based model using more than 20 million paired multiparametric measurements as ground truth. This training data linked measurements of single-cell mass, volume, density, and morphological features extracted from inline brightfield images for each individual cell. The model was trained to minimize prediction error for biophysical measurements from images alone, creating a framework where simple imaging can serve as a proxy for comprehensive biophysical profiling. The biophysical inference model accurately predicted single-cell mass (RMSE 2pg) and volume (RMSE 15fL) from brightfield images with an R2 0.95, exceeding the performance of existing gold-standard instrumentation. When deployed alongside our multiparametric platform, biophysical inference increased single-cell throughput more than 50-fold while maintaining measurement concordance. Notably, image-only predictions of immune cell activation readout, previously shown to predict neoadjuvant ICB response, achieved 95% accuracy, functionally equivalent to direct biophysical measurements in our CLIA-validated workflow. This work establishes a paradigm for inferring quantitative biophysical properties from brightfield microscopy, bridging high-content single-cell biophysics with scalable imaging workflows. Unlike existing approaches relying on large patient cohorts to correlate high-dimensional features with outcomes, our framework uses experimentally measured biophysical parameters as interpretable intermediates linking images to patient response. Millions of single-cell measurements acquired rapidly and inexpensively provide an efficient training foundation. Furthermore, the framework enables tiered deployment: core laboratories use the full multiparametric platform for measurements and model training, while peripheral sites deploy low-cost imaging for routine applications. Citation Format: Nicholas Calistri, Selim Olcum, Robert Kimmerling, Steven Wasserman, Rachel LaBella, Madeleine Vacha, Katelin Katsis, Reginald Aikins, Maria Ssozi, . A deep learning framework for extracting comprehensive single-cell biophysical profiles from brightfield images abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 5473.