BenchSci

BenchSci provides EMET, an agentic research platform for preclinical drug discovery in biopharma. Researchers ask a research question in plain language. An orchestration layer breaks the question into steps, selects scientific skills, and runs them across curated data sources, models, software packages, and code execution. A knowledge graph then verifies the results, and EMET returns cited insights, reports, and visualizations. Each result links to its underlying evidence.

Workflows

EMET provides purpose-built agent workflows for preclinical R&D functions, including:

• Discovery biology: disease mechanisms, pathways, and target validation evidence.
• Small molecule discovery: hit identification, structure–activity relationship (SAR) analysis, and compound optimization.
• Large molecule and biologics: antibody design, epitope mapping, and developability profiling.
• Translational and biomarker work: biomarker candidates and companion diagnostics.
• Drug metabolism and pharmacokinetics (DMPK): absorption, distribution, metabolism, and excretion (ADME) and pharmacokinetic (PK) predictions.
• Comparative biology and safety: off-target risks and cross-species toxicities before in vivo studies.
• Quantitative biology and pharmacology, instrumental analytics, data science, portfolio management and research operations, and regulatory work: these functions are also covered, including preclinical evidence packages and submission-ready documentation.

Architecture and Data

EMET is organized in four layers: interfaces, scientific agents, enterprise security and governance, and a ground-truth and reasoning layer. The interfaces are the EMET application, programmatic access, and lab-in-the-loop connections, including to self-driving labs. The agents are BenchSci's agents plus custom agents and workflows built per user. The reasoning layer combines scientific skills, tools, models, and science-led quality assurance and evaluation.

The data layer draws on more than 38 million scientific publications, including 16 million closed-access papers available through publisher partnerships and licenses. It also covers patents, preprints, omics databases, clinical trials, ontologies, and reagent and model-system data. A proprietary knowledge graph with 2.2 billion relationship edges grounds and verifies results. Customers can add proprietary internal data.

Models and Deployment

EMET is model-agnostic. It routes questions across multiple frontier large language models in parallel and also uses specialized models and tools such as ESM-2, AbLang2, and RDKit. Deployments include a dedicated scientific team that provides training and builds custom connectors.