Sihan Chen

• Strategic Alignment: Partner directly with the CTO and MLE team to translate "brutal" drug-discovery problems into actionable ML architecture improvements. • Model Performance & Reliability: Own the scientific evaluation framework for Neo, ensuring the foundation model meets rigorous drug-discovery requirements for complex protein-protein interfaces (PPI). • Data Strategy & Automation: Direct the acquisition and curation of high-fidelity chemical datasets, implementing automated curation pipelines to scale training inputs.

• Lead a nine-member multidisciplinary team spanning structure-based design, cheminformatics, and QSAR modeling to advance AI-driven small-molecule discovery across internal and partnered programs. • Integrated Neo-1 (LLM-based co-folding model) and NeoMol (geometric deep learning molecule generator) into drug-discovery pipelines; developed new features and data-feedback modules to enhance model utility in virtual screening. • Combined molecular dynamics, ABFE/RBFE, and ML scoring functions for multi-objective optimization of potency, permeability, and degradation efficiency. • Collaborated with medicinal chemists, biologists, and ML engineers to establish end-to-end design loops linking modeling with experimental validation

• Contributed to drug-target interaction modeling, SAR analysis, and toxicity predictions. • Applied AI- and physics-based methods for COVID-19 compound repurposing using structure-based modeling. • Built automation pipelines for molecular dynamics simulations, virtual screening, and cheminformatics workflows, enabling reproducible and scalable model evaluations across multiple discovery projects.

• Applying an information maximizing variational auto-encoder(InfoVAE) to create artificial protein folding trajectories from a continuous latent space. • Investigating thermodynamics and kinetics of protein-ligand association/dissociation using a novel free energy algorithm - Milestoning Method. This powerful algorithm provides both free energy profile and rate constant of ligand binding/unbinding simultaneously. • Using unsupervised learning algorithms, e.g. PCA, ICA, Isomap, t-SNE, and autoencoder to reduce the dimensionality of all-atomic simulation trajectories (typically 1000 - 10,000 original dimensions). • Constructed QSAR and calssification prediction model for pharmacore search and ranking (Random Forest and Navie Bayes). • Clustering (K-means and DBSCAN) millions of protein structures using important features at the active pocket, including residue RMSD, Cartesian coordinates, or principal components. • Used SVM to extract the most dissimilar motions of an enzyme before and after binding by an inhibitor.

• 2+ years of TA in graduate-level statistical thermodynamics, advanced quantum mechanics, and molecular dynamics simulations. • Wrote several tutorials and scripts for running simulations on supercomputers. • 2 years of TA in undergraduate physical chemistry • 1+ years of TA in general chemistry laboratory • Instructed and supervised 100+ undergraduates' lab skills and safety • Mentor of Ohio supercomputer center summer camp for high school students.

• Built Molecular Dynamics (MD) simulations on fluidics in nanopores, peptide adsorption on silica nanoparticles, and surface potential at interfaces. • Developed C++ programs to calculate second harmonic generation (SHG) signals at MD trajectories. • Implanted OpenMP and CUDA to in-house programs to achieve high-performance computing on supercomputers. • Performed nonlinear regression to partial differential equations, e.g. Poisson-Boltzmann, Navier-Stokes using Mathematica, Matlab, and Python (numpy).

• Developed a nanoparticle-based biosensor to detect enzymatic activity of phospholipase A2 (PLA2). • Invented a protocol to synthesized lipid bilayer coated nanoparticles, with high stability in salted environments.

• Invented a Surface-Enhanced Raman Scattering (SERS) nanoprobe, which amplifies the Raman scattering signals of naphthalenethiol molecules by 3 orders. • Cooperated with colleagues in National Central University in fabricating 2D waveguide as multiple spots sensor chips. • Performed nanoparticle characterizations using size exclusion chromatography (SEC), scanning electron microscopy (SEM), transmission electron microscopy (TEM), and dynamic light scattering (DLS).

Dr. Sherwin Singer's Team