InstructPro2: Natural Language and Biological Grounding Guided De Novo Protein Design

Charles Novak; Zhenqiao Song; Chuan Li; Jianwen Xie; Lei Li

cs.LGOct 14, 2026

InstructPro2: Natural Language and Biological Grounding Guided De Novo Protein Design

Authors: Charles Novak, Zhenqiao Song, Chuan Li, Jianwen Xie, Lei Li

Organizations: Ray and Stephanie Lane Computational Biology Department, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA · Language Technologies Institute, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA · Lambda Inc, San Francisco, CA 94105, USA

Abstract

The ability to generate proteins with desired functions is a central goal of computational biology. Although natural language can express diverse design objectives, existing protein generation methods often support only narrow tasks or make limited use of biological constraints. We develop InstructPro2 to translate natural-language design objectives into protein sequences while incorporating task-specific biological context. A text encoder captures the design instructions, while dedicated ligand and antigen encoders provide biological context to the sequence-generating decoder. For pretraining, we assemble InstructPro2Bench, comprising 205.7 million <function description, protein> pairs from UniProtKB. We subsequently adapt the model to ligand-binding protein and antibody design. On functional protein generation benchmarks, InstructPro2 outperforms Pinal, the strongest baseline, achieving a functional recovery F1 of 74.8 versus 62.2 and lowering global RMSD from 5.33A to 3.04A. For antibody design, the model obtains a DockQ score of 0.39, while ligand-binding protein design reaches a docking success rate of 72.6%. These findings highlight how integrating language-based instructions with biological conditioning can extend protein generation to distinct design settings.

Figures & tables

Figure 1: Overview of the InstructPro2 architecture. (a) During pretraining : natural-language function descriptions are encoded by the text encoder, and then the protein decoder takes the encoded representations as prefix to generate a consistent protein sequence. (b) During fine-tuning: biological grounding is incorporated alongside textual representations to enable contextually realistic protein design for real-world applications, such as ligand binding and antibody engineering.

Figure 2: Automated pipeline to construct InstructPro2Bench

Model	pLDDT $\uparrow$	Gl. RMSD (Å) $\downarrow$	Core RMSD (Å) $\downarrow$	Gl. $<$ 2Å (%) $\uparrow$	Core $<$ 2Å (%) $\uparrow$	F1 $\uparrow$
ProGen2 (2.7B)	68.49	7.80	6.05	43.5	47.4	53.3
ESM3 (1.4B)	74.51	10.78	7.65	23.3	31.8	40.0
ESM3 (7B)	77.06	8.93	5.76	29.6	45.7	52.5
Pinal (1B)	86.25	5.33	5.49	33.5	52.0	62.2
InstructPro2	84.48	3.04	2.69	63.5	67.3	74.8

Table 1: Pretrained model performance on InstructPro2Bench. InstructPro2 achieves the best performance across all the function evaluation and structure accuracy metrics, showing its strong capability of designing proteins following human instructions.

Model	pLDDT $\uparrow$	ipTM $\uparrow$	DockQ $\uparrow$
Ground Truth	87.92	0.62	–
ESM3	86.47	0.50	0.29
Pinal	78.65	0.38	0.17
ProGen2	82.35	0.49	0.28
RefineGNN	85.60	0.52	0.38
InstructPro2-Ab	86.48	0.55	0.39

Table 2: Model performance on antibody design task.

Model	pLDDT $\uparrow$	ipTM $\uparrow$	Pocket RMSD (Å) $\downarrow$	Succ $<$ 2Å (%) $\uparrow$	GNINA Affinity $\downarrow$
Ground Truth	89.63	0.92	–	–	-7.35
ESM3	46.68	0.29	12.33	1.2	-0.13
Pinal	84.89	0.89	4.21	53.9	-6.32
ProGen2	82.76	0.80	4.14	66.5	-6.96
Proteina-complexa	73.58	0.62	11.59	0.1	-5.25
InstructPro-3B	89.09	0.90	3.40	67.9	-7.64

Table 3: Ligand-binding protein design, seen ligands. InstructPro2-Lb places the binding pocket most accurately and succeeds on the most targets. All models are evaluated on the 1,288 targets completed by every method.

Model	pLDDT $\uparrow$	ipTM $\uparrow$	Pocket RMSD (Å) $\downarrow$	Succ $<$ 2Å (%) $\uparrow$	GNINA Affinity $\downarrow$
Ground Truth	89.01	0.92	–	–	-4.48
ESM3	51.01	0.36	10.85	0.0	-0.01
Pinal	90.51	0.90	2.56	78.3	-4.47
ProGen2	86.12	0.85	2.47	81.5	-4.45
Proteina-complexa	85.09	0.70	9.38	0.0	-2.73
InstructPro-3B	90.33	0.88	3.05	70.0	-4.81

Table 4: Ligand-binding protein design for unseen ligands, testing generalization to ligands not observed during training. InstructPro2-Lb places the binding pocket most accurately while matching the best baseline on success rate. All models are evaluated on the 126 targets completed by every method.

Figure 3: Analysis of decoding strategy and modality conditioning. (a) Greedy decoding against nucleus sampling (

p=0.95

\tau=0.8

) on InstructPro2Bench. (b) InstructPro2-Ab and (c–d) InstructPro2-Lb, each with and without the auxiliary modality encoder.

Figure 4: Case studies of the fine-tuned InstructPro2 designs. Designed structures are shown in blue, ground-truth structures in orange, and the antigen in gray. Full prompts are provided in Appendix D . Both cases show strong binding to the binding partner, demonstrating InstructPro2’s ability to leverage both human instructions and biological context to design functional proteins.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Component	Examples	Description
Primary (SAbDab + blueprints)	4,907	Train 4,030 / Val 433 / Test 444
Heavy-chain-only supplement	13,407	Template descriptions; of 15,124 processed
Heavy+antigen supplement	3,287	Post CDR-H3 dedup, from 5,316 raw
Total	21,601

Appendix

Table 5: Composition of the Stage 2 antibody corpus.

Component	Parameters	S1	S2
ProGen2-large-BFD90 backbone (32 layers, $d{=}2560$ )	2,648,232,992	✓	✓
Protein LM head	131,123,200	✓	✓
Text encoder (PubMedBERT, first 3 layers)	45,102,336	✓	frozen
Text projection $W_{p}$ ( $768\rightarrow 2560$ )	1,968,640	✓	✓
Text reconstruction head (3 layers + LM head)	46,704,956	✓	✓
Gated cross-attention adapters ( $8\times$ )	209,838,088	–	✓

Appendix

Table 6: Component-level parameter counts, measured from the released Stage 2 checkpoint. “S1” and “S2” indicate whether a component is updated in Stage 1 (pretraining) and Stage 2 (finetuning). The adapters and modality projections are introduced in Stage 2 and therefore do not exist during Stage 1; the Stage 1 model accordingly contains 2,990,576,029 parameters, of which 2,873,132,124 are trainable.

Figure 5: Additional case studies of InstructPro2-Ab or InstructPro2-Lb designs. Each row shows the natural-language input prompt (left), the designed structure superimposed on the ground-truth structure (middle), and a close-up of the binding region (right). Designed structures are shown in blue and ground-truth structures in orange; the antigen is shown in grey in the antibody designs. Ligand-binding designs are marked as drawn from the seen or unseen evaluation split. Ligand prompts are abbreviated (…) for space, with complete descriptions in Appendix D .

Explore similar work

Designing Protein Structures from Natural Language
Oct 9, 2026Joseph Lee, Yanjing Li, Bo Qiang +7Protein Design Generative Modeling
University of Washington · National University of Singapore · Allen Institute for AI (Ai2) +1
Proteo-R1: Reasoning Foundation Models for De Novo Protein Design
May 1, 2026Fang Wu, Weihao Xuan, Heli Qi +26Generative Modeling 3D Molecular Generation
Stanford University
AMix-2: Establishing Protein as a Native Modality in Large Language Models
May 29, 2026Keyue Qiu, Yixin Wu, Lihao Wang +19Multimodal Large Language Models Protein Language Models
1Shanghai Artificial Intelligence Laboratory · Institute for AI Industry Research (AIR), Tsinghua University

Oct 9, 2026J/K move · Enter open · S save

Designing Protein Structures from Natural Language

Joseph Lee, Yanjing Li, Bo Qiang, Gregory Lau+6

University of Washington · National University of Singapore · Allen Institute for AI (Ai2)+1

Protein Design Generative Modeling Protein Binder Design

Open paper AskPDF arXiv

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

K_{\mathrm{D}}

of 107 nM. These results establish natural language as an effective and powerful conditioning interface for protein structure generation.

Abstract

Explore similar work

Designing Protein Structures from Natural Language

Proteo-R1: Reasoning Foundation Models for De Novo Protein Design

AMix-2: Establishing Protein as a Native Modality in Large Language Models