Building a working mental model of a protein typically requires weeks of reading, cross-referencing crystal and predicted structures, and inspecting ligand complexes, an effort that is slow, unevenly accessible, and often requires specialized computational skills. We introduce Speak to a Protein, a new capability that turns protein analysis into an interactive, multimodal dialogue with an expert co-scientist. The AI system retrieves and synthesizes relevant literature, structures, and ligand data; grounds answers in a live 3D scene; and can highlight, annotate, manipulate and see the visualization. It also generates and runs code when needed, explaining results in both text and graphics. We demonstrate these capabilities on relevant proteins, posing questions about binding pockets, conformational changes, or structure-activity relationships to test ideas in real time. Speak to a Protein reduces the time from question to evidence, lowers the barrier to advanced structural analysis, and enables hypothesis generation by tightly coupling language, code, and 3D structures. Speak to a Protein is freely accessible at https://open.playmolecule.org.
Figures & tables
Figure 1: Overview of the system architecture. The Frontend hosts a chat panel, a 3D molecular viewer, and a Python console. The Agent Harness runs the agent loop, holds conversation memory, and executes tool calls. A search service indexes codebases, documentation, and literature. A sandboxed filesystem provides a per-project workspace and a preinstalled scientific software stack, with restricted network egress.
Group
Tool
Function
Sandbox
run_bash
Runs a bash command in the container.
cancel_task
Stops a background command.
Files
read_file
Reads a text file.
write_file
Writes a file.
edit_file
Replaces text inside a file.
view_image
Returns an image to the agent.
Table 1: Tools available to the agent. The agent calls each tool by name. Sandbox and file tools act on the per-project workspace; viewer tools act on the 3D scene in the frontend through the session WebSocket.
Figure 2: Interactive analysis of the CDK2 structure (PDB: 1AQ1) using the Speak to a Protein multimodal assistant. The user enters natural language queries in the chat panel (right), and the system responds with both textual answers and real-time updates to the 3D visualization (left). Python code executed in the viewer’s browser Python session is shown in the console (bottom), providing full transparency and reproducibility of the underlying analyses.
Figure 3: Accuracy on BixBench-Verified-50. Baseline scores are reproduced from the public leaderboard; error bars for Speak to a Protein indicate variability across repeated runs.
Figure 4: Task accuracy and cost on Common Protein Analysis Tasks , by group and ablation. Bars are the mean over three seeds and error bars their standard deviation. The visualization group has no bar for the ablations without a viewer, because those tasks cannot be attempted without it.
Figure 5: Structural workflow of the D3R/D2R selectivity case study. ( 5(a) ) D2R (PDB 6CM4, risperidone-bound, blue) overlaid on D3R (PDB 3PBL, eticlopride-bound, orange) after C α alignment, with the orthosteric ligands as licorice. ( 5(b) ) Close-up of the orthosteric and adjacent secondary pocket; selectivity-relevant residues are labelled for D2R (cyan) and D3R (yellow), including the conserved orthosteric Asp 3.32 anchor (D2R Asp114/D3R Asp110). ( 5(c) ) The Speak to a Protein interface at the end of the case study, showing the curated CSV of D3R-selective ligands (left) and the chat panel (right).
Figure 6: Figures the agent generated and embedded in its own report. (a) Annotated 3D-viewer snapshot of the cyclin E-dependent allosteric interface pocket (PDB 8VQ3), jointly framed by CDK2 (green) and cyclin E1 (blue) residues, captured by the agent during the analysis. (b) Matched-compound CDK2 selectivity relative to CDK1, CDK4, CDK6, CDK7, and CDK9 from filtered ChEMBL bioactivity; the dashed line is parity, the dotted red line marks a one-log margin in favor of CDK2.
A central challenge in de novo protein design is generating proteins with desired structural and functional properties. Existing design models typically require a dedicated input representation for each property, limiting the design objectives they support. We develop Prompteina, the first all-atom structure generation model that supports new and broader design objectives specified in natural language, analogous to text-to-image generation. To train Prompteina, we construct Prompteina-Data, a paired text--protein structure dataset of 1.8M structures and 50M+ captions describing global and residue-specific structural and functional properties. We adapt Proteina-Complexa, a pretrained structure generation model, through text-conditioned pretraining and finetuning that aligns residues with their corresponding text. Prompteina outperforms baselines on 15 of 18 structural properties and generates proteins that more closely match descriptions of their intended functions. In binder design, it leads baselines on at least 13 of 20 targets and improves interface confidence by 14% over the strongest baseline. With iterative in silico feedback, Prompteina's gain over non-iterative generation is 57% larger than the next-best baseline on the hardest targets. Finally, we experimentally validate Prompteina designs against GM2A, a target with no reported de novo binders. We identify nine binders, with the strongest at an apparent KD of 107 nM. These results establish natural language as an effective and powerful conditioning interface for protein structure generation.
Joseph Lee, Yanjing Li, Bo Qiang +7
University of Washington · National University of Singapore · Allen Institute for AI (Ai2) +1
Protein-language systems are often evaluated by whether they generate plausible biological text, but a structural question has a sharper semantics: it denotes a measurement in a 3D coordinate system. We introduce ProtStructQA, an executable benchmark for protein structural question answering in which each natural-language question is generated from a hidden typed domain-specific language (DSL) program and the answer is obtained by executing that program on an AlphaFold-predicted structure. ProtStructQA releases 382.2K questions covering confidence, distances, predicted aligned error (PAE), solvent exposure, secondary structure, topology and contacts, and held-out compositions: a 330K active benchmark over 10K proteins from four species, plus a 52.2K hard-negative robustness pool. Without fine-tuning, we evaluate Qwen3 models from 0.6B to 8B under direct prompting, chain-of-thought, grammar-constrained executable voting, executable voting with chain-of-thought, and multi-turn ReAct-style tool use, and replicate the headline finding on Gemma-3-1B and Gemma-3-12B. We find a capability-dependent denotation threshold between Qwen3-1.7B and Qwen3-4B: below it, tool-mediated ReAct dominates because models often fail to produce executable denotations; above it, chain-of-thought flips from mostly harmful to strongly beneficial and becomes the strongest strategy on most splits. Parse-failure and family-level analyses show that the threshold is a transition from unparseable language to executable structural denotation, while grammar and execution remain selectively valuable for PAE and secondary-structure queries. ProtStructQA reframes scientific QA as compilation from language to measurement and provides a diagnostic testbed for when language models can map words to executable 3D structural measurements.
We present AMix-2, a protein-text foundation model that establishes protein as a native modality in large language models (LLMs), unifying protein understanding and sequence design within a single foundation model. AMix-2 is built upon two key ideas: (1) a unified protein-text formulation that embeds natural language and protein sequence in a shared token space, enabling one model to perform biological reasoning and conditional design instead of separate downstream task-specialized models; and (2) a block-wise diffusion language modeling backbone that combines causal generation across blocks with bidirectional context and iterative refinement within blocks. This scheme better matches the intrinsic nature of proteins than a strict left-to-right factorization. To evaluate protein foundation models under realistic generalization settings, we further introduce ProteinArena, a comprehensive benchmark with time-aware and homology-aware protocols across various understanding and design tasks, and with baselines covering classical bioinformatics tools, protein-specialized models and LLMs. On ProteinArena, AMix-2 outperforms frontier LLMs and demonstrates competitive performance to task-specific protein models. Controlled experiments further show that the diffusion-based paradigm generally surpasses its autoregressive counterpart, highlighting the advantage of flexible generation order for protein sequences. We release both AMix-2 and ProteinArena to facilitate open research in protein foundation models.
Keyue Qiu, Yixin Wu, Lihao Wang +19
1Shanghai Artificial Intelligence Laboratory · Institute for AI Industry Research (AIR), Tsinghua University