GRASP: Graph Representation Learning with Assay Supervision for Molecular Properties
- 1Shiv Nadar University
Abstract
Molecular structures are abundant, while experimental bioactivity is sparse and distributed across assays. We introduce GRASP, a 93.5M-parameter graph Transformer that learns from these sources in sequence. GRASP first learns molecular structure through replaced-token detection on 1.54 billion ZINC20 molecule presentations, then adapts to sparse bioactivity across 642 ChEMBL assays. Across 23 cluster-held-out OpenADMET endpoints, GRASP achieves a mean MAE of 0.374, compared with 0.383 for CheMeleon, 0.408 for ECFP4-LightGBM, and 0.440 for Chemprop. ChEMBL adaptation reduces mean MAE from 0.407 to 0.374 and improves 21 of 23 endpoints, showing the value of intermediate bioactivity supervision. On 22 TDC ADMET tasks, GRASP gives a better point estimate than CheMeleon on 13 tasks. These results support staged molecular representation learning, where abundant molecular structure and sparse experimental measurements provide complementary supervision at different stages of training.
Related
- Neural encoding of grasp and object properties in the posterior parietal and motor cortices of the cortical grasping network in tetraplegic humans
- Tri-Modality Representation Learning for Molecular Property Prediction
- DECIPHER integrates disentangled representation learning and prototype-based cell-type deconvolution across molecular modalities
- Xenograft supervision and joint methylation-fragment-end encoding define a reusable molecular scoring layer for cell-free tumor DNA
- Foveation as self-supervision signal: Blur masks beat blank masks for learning robust visual representations
BibTeX
@misc{srivastava2026grasp,
title = {{GRASP: Graph Representation Learning with Assay Supervision for Molecular Properties}},
author = {S. P. Srivastava and R. Gorantla and H. Singh and S. K. Chundru and A. Mey and R. K. SINGH},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.28.755019},
url = {https://www.biorxiv.org/content/10.64898/2026.09.28.755019v2},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - GRASP: Graph Representation Learning with Assay Supervision for Molecular Properties
AU - S. P. Srivastava
AU - R. Gorantla
AU - H. Singh
AU - S. K. Chundru
AU - A. Mey
AU - R. K. SINGH
PY - 2026
DA - 2026/09/28
PB - bioRxiv
DO - 10.64898/2026.09.28.755019
UR - https://www.biorxiv.org/content/10.64898/2026.09.28.755019v2
AB - Molecular structures are abundant, while experimental bioactivity is sparse and distributed across assays. We introduce GRASP, a 93.5M-parameter graph Transformer that learns from these sources in sequence. GRASP first learns molecular structure through replaced-token detection on 1.54 billion ZINC20 molecule presentations, then adapts to sparse bioactivity across 642 ChEMBL assays. Across 23 cluster-held-out OpenADMET endpoints, GRASP achieves a mean MAE of 0.374, compared with 0.383 for CheMeleon, 0.408 for ECFP4-LightGBM, and 0.440 for Chemprop. ChEMBL adaptation reduces mean MAE from 0.407 to 0.374 and improves 21 of 23 endpoints, showing the value of intermediate bioactivity supervision. On 22 TDC ADMET tasks, GRASP gives a better point estimate than CheMeleon on 13 tasks. These results support staged molecular representation learning, where abundant molecular structure and sparse experimental measurements provide complementary supervision at different stages of training.
ER -