Benjamin Sánchez-Lengeling

Introduces TasteBench, a multimodal benchmark and competition for sensory prediction across 21K+ human evaluations on 215 plant-based foods and 15K flavor molecules, providing computational proxies to accelerate sustainable food discovery.

Couples deep learning turnover number predictions with stochastic simulated annealing to parameterize enzyme-constrained genome-scale models across 93 organisms, resolving long-standing biocatalytic data sparsity.

Crowdsources predictive models across 26 international teams in the DREAM Olfactory Challenge, demonstrating that compact semantic representations of single odorants generalize to complex mixture perceptual spaces.

Comprehensive review articulating an AI-driven framework for sustainable food formulation, spanning perception modeling, multi-omics fermentation engineering, and consumer acceptance.

Develops differentiable heterogeneous graph neural network surrogates for kinetic Monte Carlo simulations of core-shell upconverting nanoparticles, achieving 6.5-fold experimental emission enhancement via gradient ascent.

Pairs a GINE molecular graph encoder with frozen ESM-C protein language model embeddings via cross-attention to predict covalent target engagement across pathogen proteomes with few-shot transfer.

Establishes the first unified open-source benchmark suite spanning 11 diverse chemical mixture domains with standardized splitting protocols and deep learning baseline models.

A foundational pedagogical text and monograph introducing structural, geometric, and kinetic graph neural networks for molecules, macromolecular proteins, and complex biological pathways.

Introduces POMMix, incorporating inductive perceptual pooling operations on molecular graphs to predict the non-linear emergent sensory properties of multi-component odorant mixtures.

Demonstrates that pairwise ranking surrogates systematically outperform point regression models in Bayesian optimization for molecular property design, mitigating miscalibration and heteroskedastic noise.

Leverages graph neural representations from the Principal Odor Map (POM) to discover potent, novel repellent chemotypes validated experimentally against multiple mosquito disease vectors.

Introduces Stereoelectronics-Infused Molecular Graphs (SIMGs) and neural surrogate estimators that encode Natural Bond Orbital interactions into molecular graphs in seconds, boosting property prediction accuracy.

A globally distributed, asynchronous self-driving laboratory discovers 21 state-of-the-art organic solid-state laser gain materials across multiple sites and automated robotics testbeds.

Synthesizes key outcomes from the AI4Mat workshop at NeurIPS 2023, charting trajectories for self-driving laboratories, simulation-to-materials workflows, and language models for physical science.