Claude Protein Design and Analytical Chemistry Capabilities

TL;DR

Anthropic has demonstrated that Claude models can significantly accelerate life science research by autonomously designing high-affinity protein binders and automating the analysis of raw chemical data. In protein design, Claude achieved hit rates between 22% and 35%—surpassing the industry typical of 10-15%—while in analytical chemistry, Claude Opus 5 processed raw NMR and LC-MS files in under 25 minutes, matching the accuracy of professional contract labs.

Autonomous Protein Binder Design

Claude models can execute de novo protein design campaigns with minimal human intervention, producing binders that often match or exceed the affinity of the best previously published results.

Performance and Hit Rates

Using a combination of Claude Opus 4.8 and Mythos Preview, Anthropic tested the models against 15 targets. The models successfully designed binders against 14 of those targets.

Hit Rates: In a multi-target 48-hour session, Mythos Preview and Opus 4.8 achieved overall hit rates of 26.7% and 22.6%, respectively. When Mythos Preview focused on targets individually in 24-hour sessions, the hit rate increased to 35.1%.
Comparison: These results significantly outperform the typical 10-15% hit rate seen in current protein design campaigns.
Affinity: Claude produced high-affinity binders (defined as $K_D < 10$ nM) for at least six targets and matched or exceeded the best reported affinity for at least four targets.

Target-Specific Successes and Limitations

Claude demonstrated specialized capabilities across different protein structures:

RBX1: Mythos Preview (single-target mode) achieved a 40% hit rate, compared to 3.7% among human participants in an Adaptyv Bio competition, with a top design that outperformed the winning human entry.
TNFα: While Mythos Preview failed, Opus 4.8 successfully designed binders for this challenging multimeric target, including species cross-reactive binders that bound human, cynomolgus monkey, and mouse TNFα.
β-sheets: Claude designed 15 confirmed binders containing at least 20% β-strand across six targets, showing an ability to reason about complex secondary structures beyond standard α-helices.
Challenges: Claude struggled with Maltose Binding Protein (MBP) due to its smooth surface and had limited success with BBF-14, a de novo designed β-barrel protein.

Technical Execution

Claude operated autonomously within "Claude Science," orchestrating existing open-source structure design, sequence design, and co-folding models. The model handled the entire pipeline: selecting target sites, generating candidate structures, performing in silico optimization, and screening for solubility and expression. The only human involvement was granting network/code execution approvals and ordering the physical designs for wet lab validation by Adaptyv Bio and Twist Bioscience.

Automation of Analytical Chemistry

Claude Opus 5 can automate the time-intensive process of interpreting raw instrument data used to verify the identity and purity of chemical compounds.

NMR and LC-MS Processing

In a test using raw files from a contract lab and a two-sentence prompt, Claude Opus 5 processed the data in parallel:

NMR Spectroscopy: Claude converted raw signal (free induction decay) into a calibrated spectrum and a table of 18 peaks. It identified four broad peaks as likely nitrogen or oxygen attachments and correctly proposed a heavy-water swap experiment to verify them. Its hydrogen counts per peak were within 0.08 ¹H of the lab's results.
LC-MS (Liquid Chromatography-Mass Spectrometry): Claude decoded an undocumented vendor binary format to extract the chromatogram and mass spectra. It measured the compound purity at 96.4%, nearly identical to the lab's 96.33%.

Efficiency Gains

Claude completed both analyses in 23 and 19 minutes, respectively. In contrast, the contract lab's finished report for the same sample arrived four days after the initial spectrum was acquired, highlighting a significant reduction in the lag time associated with manual human analysis.

Safety and Dual-Use Considerations

Anthropic notes that agentic biological discovery is "dual-use," meaning the same capabilities that accelerate therapies could potentially be used to develop bioweapons. Consequently, protein design and other dual-use biology capabilities remain unavailable for general access in Claude Fable 5. Anthropic is developing trusted access programs to provide these tools to scientists safely.