The AI for Scientific Discovery Summit, presented by SandboxAQ, brings together customers, partners, and scientific leaders across BioSim and ChemSim to explore how AI is accelerating discovery across drug discovery, chemistry, biology, and materials innovation.
Through customer stories, product insights, and peer discussion, the event will highlight how physics-grounded AI and large quantitative models are helping research teams move faster, make better decisions, and advance scientific progress across both molecular and materials R&D.
Brings nearly 40 years of experience in organizational development and technology introduction, including more than 15 years in life sciences. With prior experience at Biogen, the Broad Institute, and Bristol Myers Squibb, along with work advising on commercialization, knowledge graphs, customer community building, and strategic partnerships.
Volker Eyrich is a Customer Engineer on Google Cloud’s Healthcare & Lifesciences team where he focuses on the areas of Scientific Research, Scientific and High Performance Computing and AI Infrastructure. He joined Google in 2018.
Prior to Google he spent 18 years in Computational Drug Discovery focusing on algorithm and infrastructure development. In that context he used Google Cloud extensively for computationally intensive calculations as well as scalable compute and data analytics infrastructure.
Leading commercial strategy and operations at Apheris, an AI drug discovery company working with 15+ pharmaceutical companies, nine of them top-20 pharma. Apheris customizes and fine-tunes discovery models on the proprietary data locked inside pharma, which only a neutral federated network is allowed to reach, spanning co-folding, binding affinity and ADMET, so that predictions become accurate enough to change decisions in live drug programs.
Specializing in knowledge graph construction and machine learning for drug discovery, leading development of the AI Sim knowledge graph platform and cheminformatics/ML tooling (Shortpath, ModelForge) that power target identification and biomarker discovery efforts like the MJFF Parkinson's disease program
Chemist and machine learning engineer, he develops AI/ML, computational chemistry, and software applications that accelerate drug discovery. His work spans the pipeline from target identification and virtual screening through hit-to-lead, lead optimization, ADMET/DMPK profiling, and drug development, with experience in small molecules, targeted protein degradation, and antibody engineering. He has contributed to more than 10 small-molecule and antibody programs in collaboration with academic, biotech, and pharmaceutical partners.
Product leader on SandboxAQ’s AI Simulation team, focused on Large Quantitative Models for healthcare, life sciences, and chemistry. She brings experience in computational chemistry, quantum computing, high-performance computing, and scientific product development from Microsoft and AWS. She holds a Ph.D. in Computational Chemistry from the University of Washington.
Leads product and ecosystem strategy for SandboxAQ’s core scientific AI assets, transforming proprietary Large Quantitative Models (LQMs) and physics-based datasets into enterprise-grade discovery tools deployed via AWS, GCP, Azure, and MCP. Directs a cross-division portfolio spanning drug and materials discovery, backed by more than a decade of work across life sciences and advanced materials.
Works at the intersection of AI, physics-based simulation, and materials science to accelerate the discovery and commercialization of next-generation semiconductor materials. Her work spans critical innovation areas including catalysts, PFAS-free process chemicals, magnets, and battery systems, and she has spoken publicly on AI-accelerated materials engineering and digital tools for operational efficiency.
Currently, drug discovery teams using physics-based virtual screening drug discovery teams spend weeks managing GPU infrastructure and writing orchestration code. SandboxAQ connects Large Language Models to Large Quantitative Models via the Model Context Protocol to eliminate this friction. You can execute complex computational workflows using plain English. During this session, we will demonstrate AQPotency, our CPU-based potency prediction ranking engine. We will rank 10,000 protein-ligand pairs in a minute and reverse-screen curated panels for off-target safety signals. You prioritize drug candidates and advance therapies to market without specialized engineering support.
Co-folding and affinity models perform well on targets that resemble their training data and poorly on the ones a discovery program actually starts from. That gap is set by data coverage, not by architecture.
Adam Lewis (SandboxAQ) opens on structure-modelling at SandboxAQ, including a first look into our next generation of deep learned affinity research. Julian Schönauer (Apheris) follows with the two data strategies that widen a model's applicability domain and accuracy within drug programs: local fine-tuning on a company's own program data, and federated training across several organizations' proprietary data.
High-fidelity simulations like Density Functional Theory (DFT) drive modern R&D but are bottlenecked by complex coding and massive infrastructure demands. By integrating Large Quantitative Models (LQMs) directly into conversational AI like Anthropic's Claude, we remove this barrier entirely. During this session, we will demonstrate the LLM-to-LQM pipeline using AQCat Adsorption, a spin-aware machine learning engine that delivers near-DFT accuracy at up to 20,000 times the speed of traditional methods. We will show how translating plain-English prompts into automated computational workflows empowers your non-computational bench scientists to independently screen candidates, slash computational overhead, and accelerate physical breakthroughs without writing a single line of code.
If you build scientific software, the ceiling on your users' screening campaigns is set by the compute layer beneath your models. This session covers CUDA-X libraries you can build on directly: NVIDIA ALCHEMI, whose batched geometry relaxation and molecular dynamics have delivered up to 100x on materials stability workloads; cuEquivariance, which accelerates the equivariant neural networks behind modern interatomic potentials by up to 10x end-to-end; and cuEST, bringing GPU accelerated first-principles quantum chemistry on production workloads. You'll leave knowing what tools are available today, where the real bottlenecks still sit, and how to wire these into your own stack.
SandboxAQ’s CHIPS work will showcase how physics-based AI can move scientific discovery beyond prediction and into national-scale industrial impact. In this session, Shalini Sharma will discuss how SandboxAQ is using its ReAQT platform and Large Quantitative Models to accelerate discovery, validation, and commercialization across four semiconductor-critical materials areas: PFAS-free process chemicals, catalysts, rare earth-free magnets, and battery systems. The talk will explore how AI-driven materials workflows can compress years of trial-and-error into targeted discovery campaigns, help reduce dependence on foreign-controlled supply chains, and create a path from breakthrough science to domestic manufacturing partnerships.
SandboxAQ is a science-first technology company spun out of Alphabet in 2022. Our drug discovery team includes a dedicated biopharma core of 70+ specialists. These include over 48 domain PhDs across computational chemistry, medicinal chemistry, computational biology, bioinformatics– who work alongside AI and software engineers specialized in cloud-scale molecular simulation.
