ProtRLSearch: A Multi-Round Multimodal Protein Search Agent with Large Language Models Trained via Reinforcement Learning

Congying Liu; Taihao Li; Ming Huang; Xingyuan Wei; Peipei Liu; Yiqing Shen; Yanxu Mao; Tiehan Cui

cs.AIMar 2, 2026

ProtRLSearch: A Multi-Round Multimodal Protein Search Agent with Large Language Models Trained via Reinforcement Learning

Authors: Congying Liu, Taihao Li, Ming Huang, Xingyuan Wei, Peipei Liu, Yiqing Shen, Yanxu Mao, Tiehan Cui

Organizations: University of Chinese Academy of Sciences, Beijing, China · Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, Hangzhou, China · Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China · Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China ·

Abstract

Protein analysis tasks arising in healthcare settings often require accurate reasoning under protein sequence constraints, involving tasks such as functional interpretation of disease-related variants, protein-level analysis for clinical research, and similar scenarios. To address such tasks, search agents are introduced to search protein-related information, providing support for disease-related variant analysis and protein function reasoning in protein-centric inference. However, such search agents are mostly limited to single-round, text-only modality search, which prevents the protein sequence modality from being incorporated as a multimodal input into the search decision-making process. Meanwhile, their reliance on reinforcement learning (RL) supervision that focuses solely on the final answer results in a lack of search process constraints, making deviations in keyword selection and reasoning directions difficult to identify and correct in a timely manner. To address these limitations, we propose ProtRLSearch, a multi-round protein search agent trained with multi-dimensional reward based RL, which jointly leverages protein sequence and text as multimodal inputs during real-time search to produce high quality reports. To evaluate the ability of models to integrate protein sequence information and text-based multimodal inputs in realistic protein query settings, we construct ProtMCQs, a benchmark of 3,000 multiple choice questions (MCQs) organized into three difficulty levels. The benchmark evaluates protein query tasks that range from sequence constrained reasoning about protein function and phenotype changes to comprehensive protein reasoning that integrates multi-dimensional sequence features with signal pathways and regulatory networks.

Figures & tables

Fig. 1: A comparison between single-round search methods and the proposed approach, illustrating their differences in the search process. Red annotations indicate error cases such as missing keywords, while green annotations denote correct results.

Fig. 2: The overall agent encompasses multimodal encoding, multi-round search trained via RL, and the design of multi-dimensional rewards.

Fig. 3: Overview of the constructed training dataset

Dataset	Method	Level 1			Level 2			Level 3
		Acc%	Token	Time/s	Acc%	Token	Time/s	Acc%	Token	Time/s
BioMedMCQs	Baseline	45.8 $\pm$ 0.6	321 $\pm$ 41	6.5 $\pm$ 0.7	38.7 $\pm$ 1.2	335 $\pm$ 58	6.8 $\pm$ 1.1	29.5 $\pm$ 0.9	379 $\pm$ 72	7.6 $\pm$ 1.4
	BioMedSearch [ 4 ]	73.3 $\pm$ 0.8	3029 $\pm$ 155	48.1 $\pm$ 1.5	69.7 $\pm$ 1.4	3348 $\pm$ 121	65.1 $\pm$ 0.9	62.4 $\pm$ 1.0	3987 $\pm$ 188	78.9 $\pm$ 1.6
	BioReason [ 16 ]	63.9 $\pm$ 1.7	472 $\pm$ 63	11.7 $\pm$ 1.3	56.2 $\pm$ 0.8	496 $\pm$ 71	16.7 $\pm$ 0.8	49.3 $\pm$ 1.4	528 $\pm$ 53	20.5 $\pm$ 0.6
	Search-R1 [ 6 ]	76.1 $\pm$ 0.7	2063 $\pm$ 134	19.3 $\pm$ 0.4	70.9 $\pm$ 1.3	2582 $\pm$ 98	27.1 $\pm$ 1.2	63.5 $\pm$ 1.8	2835 $\pm$ 176	32.4 $\pm$ 1.9
	ProtLLM [ 19 ]	74.8 $\pm$ 1.1	409 $\pm$ 82	9.6 $\pm$ 1.1	60.4 $\pm$ 0.9	534 $\pm$ 65	14.9 $\pm$ 1.3	57.9 $\pm$ 0.5	560 $\pm$ 103	16.2 $\pm$ 1.5

TABLE I: Performance comparison across methods on BioMedMCQs and ProtMCQs.

Dataset	Protein	RL	Level 1			Level 2			Level 3
			Acc%	Token	Time/s	Acc%	Token	Time/s	Acc%	Token	Time/s
BioMedMCQs	$\checkmark$	$\checkmark$	89.2 $\pm$ 1.4	554 $\pm$ 117	10.5 $\pm$ 0.9	75.8 $\pm$ 0.6	785 $\pm$ 84	13.4 $\pm$ 1.6	71.7 $\pm$ 1.9	873 $\pm$ 156	21.3 $\pm$ 1.7
	$\checkmark$	$\times$	76.7 $\pm$ 1.9	977 $\pm$ 148	15.4 $\pm$ 2.3	70.7 $\pm$ 0.9	1719 $\pm$ 238	20.5 $\pm$ 1.7	65.1 $\pm$ 1.8	2178 $\pm$ 327	35.6 $\pm$ 1.3
	$\times$	$\checkmark$	83.2 $\pm$ 0.6	586 $\pm$ 103	12.6 $\pm$ 1.2	72.9 $\pm$ 1.5	874 $\pm$ 129	15.3 $\pm$ 2.1	69.8 $\pm$ 0.7	910 $\pm$ 203	22.1 $\pm$ 1.8
ProtMCQs	$\checkmark$	$\checkmark$	86.9 $\pm$ 1.3	632 $\pm$ 118	15.9 $\pm$ 0.7	77.4 $\pm$ 0.8	706 $\pm$ 91	20.4 $\pm$ 1.2	72.5 $\pm$ 1.7	989 $\pm$ 153	24.3 $\pm$ 0.4
	$\checkmark$	$\times$	78.6 $\pm$ 1.7	1016 $\pm$ 175	24.7 $\pm$ 1.8	63.7 $\pm$ 1.0	1623 $\pm$ 136	22.8 $\pm$ 1.3	61.2 $\pm$ 1.4	2031 $\pm$ 164	33.4 $\pm$ 2.0

TABLE II: Performance comparison across BioMedMCQs and ProtMCQs under different pre-training configurations.

Method	MedMCQA	MedQA
BioBERT [ 25 ]	36.7	37.1
SciBERT [ 26 ]	-	39.2
MedAlpaca-7B [ 27 ]	55.2	45.8
MEDAGENTS [ 28 ]	64.1	59.3
ProtRLSearch (w/o RL)	85.8	83.9
ProtRLSearch (w/o Prot)	88.5	85.2

TABLE III: Performance comparison on MedMCQA and MedQA (Acc %).

Multimodal Search Agents

Multimodal search agents retrieve and reason across text, images, video, and other sources to answer complex questions.

See all 47 papers

Sep 24, 2026

PFArena: Benchmarking Language Models for Protein Modification

Protein modification requires navigating an immense sequence space, yet wet-lab validation remains low-throughput and costly. Although computational paradigms including protein language models (PLMs), large language models (LLMs), and LLM-based agents have shown promise in protein modification, their relative efficacy across realistic experimental decision-making settings remains unclear. To bridge this gap, we introduce PFArena, a benchmark comprising four controlled task interfaces that cover single-mutant generation and multi-mutant ranking. By providing varying levels of mutation fitness data, PFArena reflects four representative research scenarios characterized by differing degrees of prior experimental context. We assess six PLMs, six LLMs, and five LLM-based agents using complementary metrics to measure both peak and overall protein modification performance. Our evaluation reveals that model performance shifts systematically with the availability of target-specific experimental evidence: PLMs demonstrate proficiency in open-ended single-mutant generation by leveraging protein-specific priors, whereas LLMs and agents perform strongly in multi-mutant ranking, particularly when target-specific fitness data are available. Nevertheless, all model families face fundamental challenges with increasing search-space size and mutation depth. We release our code and benchmark suite to facilitate reproducible research in model-assisted protein modification.

Yawen Ouyang, Xinbo Zhang, Ziyuan Ma +15

Shanghai Artificial Intelligence Laboratory · Generative Symbolic Intelligence Lab (GenSI), Tsinghua University · Institute for AI Industry Research (AIR), Tsinghua University +1

Protein Fitness Prediction Protein Language Models LLM Agents Protein Design

Ask PDF

Aug 26, 2026

Unlocking Multimodal Protein Language Models at Inference Time

Multimodal protein language models (pLMs) learn joint protein sequence-structure distributions, and their generation performance should also depend critically on inference-time sampling strategies. Yet prior work has focused more on model training than on how inference-time strategies behave. In this paper, we establish a three-stage investigation framework to empirically study the inference design space of multimodal pLMs across three representative pLMs and four fundamental tasks. We evaluate vanilla sampling, task-specific classifier-free guidance, and reward-guided beam search on multimodal pLMs, corresponding to controls over sampling distributions, per-step logits, and parallel trajectories. Throughout the complementary advancements centered on exploration-exploitation trade-off, we (1) reveal the suboptimality of default inference protocols and identify task-oriented sampling preferences; (2) observe substantial quantitative gains across tasks, consistently boosting the upper bound performance of multimodal pLMs without updating model parameters; (3) derive conclusions about base models that differ from prior consensus.

Yi Zhou, Qipeng Wang, Yunqing Liu +3

The Hong Kong Polytechnic University · The Hong Kong University of Science and Technology (Guangzhou)

Inference-Time Search Protein Language Models Classifier-Free Guidance

Ask PDF

Jun 1, 2026

AgentPLM: Agentic Protein Language Models with Reasoning-Augmented Decoding for Protein Sequence Design

Protein language models (PLMs) are passive oracles: they generate sequences in a single forward pass with no mechanism to consult external biophysical feedback or redirect generation when a candidate violates thermodynamic or structural constraints. We introduce AgentPLM, which addresses this by equipping a pre-trained PLM with i) Reasoning-Augmented Decoding (RAD), which interleaves autoregressive generation with tool calls (ESMFold, FoldX, AutoDock Vina), and ii) Contrastive Agent Policy Optimisation (CAPO), a trajectory-level extension of direct preference optimisation that trains the policy end-to-end to learn when oracle feedback is informative rather than merely imitating high-fitness sequences. We evaluate AgentPLM on benchmark tasks spanning de novo enzyme design, antibody optimisation, thermostability, PPI interface design, and zero-shot fitness prediction with standardised oracle APIs and controlled sequence-identity splits. AgentPLM achieves state-of-the-art results with a gain in antibody top-10% hit rate over the strongest passive baseline, providing mechanistic evidence of online error correction without explicit backtracking.

Sahil Rahman, Maxx Richard Rahman

Bedford College, London, United Kingdom · Saarland University, Saarbrücken, Germany

RL for Language Models Direct Preference Optimization Protein Language Models LLM Tool Use Molecular Generation Protein Design

Ask PDF

Abstract

Multimodal Search Agents

Explore similar work

PFArena: Benchmarking Language Models for Protein Modification

Unlocking Multimodal Protein Language Models at Inference Time

AgentPLM: Agentic Protein Language Models with Reasoning-Augmented Decoding for Protein Sequence Design