TL;DR: Researchers at Sichuan University published REAPS on September 19, 2026, a receptor-aware geometric graph neural network that designs linear and macrocyclic peptide binders and produced experimentally verified active candidates at the neurokinin-3 receptor.

REAPS Reframes Peptide Design Around the Receptor Itself

Peptide therapeutics occupy a critical niche between small molecules and biologics, yet designing sequences for peptide binders remains challenging because bioactive conformations and interfacial fitness are dictated by the receptor microenvironment, according to the article titled REAPS: An All-Atom Receptor-Aware Geometric Deep Learning Framework for De Novo Design of Linear and Macrocyclic Peptide Binders published in the Journal of Chemical Information and Modeling.

The paper, published online on September 19, 2026 by the American Chemical Society, introduces the REceptor-Aware Peptide Sequence Designer, abbreviated REAPS, as a geometric graph neural network for sequence design for peptide binders, according to ACS Publications.

The central shift is to treat the receptor as a fully observable all-atom context, rather than treating peptide design as a generic inverse folding problem that relies on backbone geometry alone, according to ACS Publications.

Multidimensional evaluations show that REAPS outperforms the widely used inverse folding model ProteinMPNN in sequence recovery, structural fidelity, and interface-level biophysical metrics across both linear and macrocyclic peptide binders, according to ACS Publications.

The team further integrated REAPS with structural hallucination in a closed-loop de novo discovery pipeline, yielding peptide candidates with experimentally verified functional activity at the neurokinin-3 receptor, abbreviated NK3R, according to ACS Publications.

The work was received on June 22, 2026, revision received on August 29, 2026, accepted on September 3, 2026, published online on September 19, 2026, and published in issue on October 12, 2026, according to ACS Publications.

Who Built REAPS and Where

The authors are Yongkang Qiu, Jingxin Qiao, Lengjing Zhu, Le Du, Shengyong Yang, and Jun Zou, according to ACS Publications and the Exa library record for the DOI 10.1021/acs.jcim.6c02057.

All listed authors are affiliated with Sichuan University, according to the Exa library record for the publication.

Corresponding contacts list Email [email protected] and Email [email protected], according to ACS Publications.

The authors declare no competing financial interest, according to ACS Publications.

The publisher is the American Chemical Society, and the journal is the Journal of Chemical Information and Modeling, according to ACS Publications.

Funding support included the Innovative Drug Research and Development National Science and Technology Major Project 2025ZD1803104, the State Key Laboratory of Drug Regulatory Science Project 2025SKLDRS0348, the Sichuan Science and Technology Program 2026NSFSC1745 to J.Q., the Postdoctoral Fellowship Program of the China Postdoctoral Science Foundation GZB20250829 to J.Q., and the China Postdoctoral Science Foundation 2025M783575 to J.Q., according to ACS Publications.

The funding statement also lists the China Postdoctoral Science Foundation, State Key Laboratory of Drug Regulatory Science Project, Innovative Drug Research and Development National Science and Technology Major Project, Sichuan Science and Technology Program, and Postdoctoral Fellowship Program of the China Postdoctoral Science Foundation, according to ACS Publications.

What Problem REAPS Addresses

Current computational approaches largely treat peptide binder design as a generic inverse folding problem that relies on backbone geometry alone, overlooking the detailed physicochemical constraints of the receptor pocket, according to ACS Publications.

Peptide bioactive conformations and interfacial fitness are dictated by the receptor microenvironment, which makes sequence design challenging when the pocket context is simplified or omitted, according to ACS Publications.

The analysis summarized by The MedChem Digest article titled All-Atom Receptor-Aware Deep Learning Model Advances De Novo Peptide Binder Design describes the same distinction: REAPS models the target receptor as a fully observable, all-atom structural context, rather than relying on backbone geometry alone as typical inverse-folding methods do.

That summary notes that all-atom receptor context should, in principle, capture side-chain packing and H-bond networks that backbone-only inverse folding misses, with particular relevance for macrocycle design where conformational preorganization matters, according to The MedChem Digest.

The REAPS paper reframes sequence design for peptide binders by making the receptor pocket fully observable at the atomic level during sequence prediction, according to ACS Publications.

How REAPS Works

REAPS is described as a geometric graph neural network that reframes sequence design for peptide binders, according to ACS Publications.

The Supporting Information for the paper includes algorithmic details and pseudocode for the InterfaceAttentionBlock, plus implementation, architecture, training protocol, hyperparameters, parameter counts, and computational requirements of REAPS, according to ACS Publications.

The same Supporting Information describes computational protocol for RFpeptides-based macrocyclic peptide scaffold generation, all-atom receptor restoration, receptor truncation to binding-pocket regions, and target-specific generation settings, according to ACS Publications.

The paper's open materials include source code, model checkpoints, processed data sets, preprocessing scripts, and an example design workflow publicly available at https://github.com/Mistletoe-git/REAPS, according to ACS Publications.

The Supporting Information also includes comparison of peptide-specific sequence-design models and inference-only evaluation of CyclicMPNN on CPCore, according to ACS Publications.

Structural comparisons include complex-wise paired effect-size analyses and structural and interface-level comparisons of REAPS with ProteinMPNN and REAPS-bb, according to ACS Publications.

How REAPS Compares With ProteinMPNN

ProteinMPNN is identified in the REAPS abstract as the widely used inverse folding model used as the comparison point, according to ACS Publications.

ProteinMPNN is the model described in the 2022 Science paper titled Robust deep learning–based protein sequence design using ProteinMPNN by Dauparas et al., according to the Max Zhuang reading page and the Chinese technical explainer on CyclicMPNN.

That baseline model generates sequences for a given protein backbone and is useful for designing new proteins or improving existing designs, according to the Max Zhuang reading page.

The aidrugsearch.com knowledge base article titled How do I effectively use ProteinMPNN for sequence optimization in drug discovery workflows? describes ProteinMPNN as a message-passing neural network that aggregates information from neighboring residues, enabling the model to capture local and long-range structural dependencies without explicit physical force fields.

The same knowledge base article describes ProteinMPNN as predicting amino acid probabilities given a fixed backbone structure, separating structure prediction from sequence generation, with inference times often measured in seconds per protein rather than hours or days.

Against that baseline, multidimensional evaluations show that REAPS outperforms ProteinMPNN in sequence recovery, structural fidelity, and interface-level biophysical metrics across both linear and macrocyclic peptide binders, according to ACS Publications.

The MedChem Digest summary likewise states that in head-to-head evaluations against ProteinMPNN, REAPS showed improved sequence recovery, structural fidelity, and interface-level biophysical properties for both linear and macrocyclic peptide binders.

The Supporting Information expands that comparison with structural and interface-level comparisons of REAPS with ProteinMPNN and REAPS-bb, including complex-wise paired effect-size analyses, according to ACS Publications.

Why Linear and Macrocyclic Binders Both Matter

The REAPS framework is explicitly presented for de novo design of linear and macrocyclic peptide binders, according to ACS Publications.

Evaluations covered both linear and macrocyclic peptide binders across sequence recovery, structural fidelity, and interface-level biophysical metrics, according to ACS Publications.

The CyclicMPNN technical explainer, titled CyclicMPNN: 环肽序列设计专用模型——ProteinMPNN 的环肽微调版, provides context for why macrocycles are a distinct case, noting that cyclic peptides in the 6-14 residue range with head-to-tail covalent cyclization differ structurally from globular proteins with hundreds of residues, secondary structure, and hydrophobic cores.

That explainer notes that ProteinMPNN training data comes from the PDB, where large globular proteins dominate, while cyclic peptides represent a different structural world, according to the same technical explainer.

The MedChem Digest summary of REAPS emphasizes that the improvement over ProteinMPNN was observed for both linear and macrocyclic binders, not only one format.

REAPS therefore addresses sequence design across flexible linear peptides and conformationally constrained macrocycles within one receptor-aware framework, according to ACS Publications.

Closed-Loop Pipeline From Hallucination to Tested Peptides

The team integrated REAPS with structural hallucination in a closed-loop de novo discovery pipeline, according to ACS Publications.

The MedChem Digest summary describes that pipeline as pairing REAPS with structural hallucination to generate novel peptide candidates without requiring a known starting scaffold.

That summary states that candidates produced through this pipeline were synthesized and tested against NK3R, where they showed experimentally confirmed functional activity, according to The MedChem Digest.

The ACS abstract similarly states that the closed-loop de novo discovery pipeline yielded peptide candidates with experimentally verified functional activity at NK3R, according to ACS Publications.

The Supporting Information includes complete functional evaluation of all 11 tested NK3R peptide binder candidates, according to ACS Publications.

It also includes certificates of analysis, analytical HPLC chromatograms, reported purity values, and mass spectrometry data supporting the identity of the synthesized NK3R peptide binder candidates, according to ACS Publications.

Why NK3R Was a Demanding Test Case

NK3R is a clinically validated G protein-coupled receptor, abbreviated GPCR, expressed on KNDy neurons in the infundibular nucleus of the hypothalamus, according to the AbCellera Biologics Inc. announcement titled AbCellera Announces Presentation of Phase 2 Data for ABCL635 at the IMS World Congress on Menopause 2026, distributed via Business Wire.

The same AbCellera announcement identifies ABCL635 as an investigational antibody drug specifically targeting NK3R for non-hormonal, long-acting treatment of moderate-to-severe vasomotor symptoms, commonly known as hot flashes, due to menopause.

Neurokinin 3 receptor antagonists have recently emerged as a promising non-hormonal treatment for vasomotor symptoms in postmenopausal women, according to the Chemicalbook article titled Elinzanetant vs Fezolinetant, Treating Symptoms of Postmenopausal Women.

That Chemicalbook article identifies fezolinetant as a well-studied NK3R antagonist with significant efficacy in reducing frequency and severity of hot flashes compared with placebo, and elinzanetant as another NK3R antagonist and newly approved therapy that demonstrated superiority over placebo in individual trials.

Interest in neurokinin receptor antagonists stemmed from the discovery that neurokinin B, abbreviated NKB, is overexpressed in the hypothalamus of women who have passed through menopause, according to the Chemicalbook article.

Preclinical studies demonstrated that kisspeptin-neurokinin B-dynorphin neurons, called KNDy neurons, located in the hypothalamus express NKB, promote cutaneous vasodilatation associated with hot flashes, and participate in estrogen-dependent modulation of body temperature, according to the Chemicalbook article.

The MedChem Digest commentary on REAPS asks about hit rate and binding affinity on synthesized NK3R peptides beyond functional activity, and whether performance holds on a GPCR-like target outside training distribution, while noting that NK3R hits with confirmed functional activity clear the test of producing binders without a template scaffold.

What Is Openly Available Now

The source code, model checkpoints, processed data sets, preprocessing scripts, and an example design workflow are publicly available at https://github.com/Mistletoe-git/REAPS, according to ACS Publications.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jcim.6c02057, according to ACS Publications.

Available supporting files include a PDF with certificates of analysis, analytical HPLC chromatograms, reported purity values, and mass spectrometry data supporting the identity of synthesized NK3R peptide binder candidates, according to ACS Publications.

A second supporting PDF covers InterfaceAttentionBlock details, REAPS implementation and training protocol, comparison of peptide-specific sequence-design models, RFpeptides-based scaffold generation protocol, receptor restoration and truncation methods, paired effect-size analyses, and complete functional evaluation of all 11 tested NK3R candidates, according to ACS Publications.

The paper lists topics as Computational Drug Discovery Methods, Chemical Synthesis and Analysis, and Protein Structure and Dynamics, according to the Exa library record.

The DOI for the paper is 10.1021/acs.jcim.6c02057, according to the Exa library record and ACS Publications.

What Changes for Peptide Drug Discovery

The reported advance is not only higher sequence recovery, but improved structural fidelity and interface-level biophysical metrics for binders that must fit a specific pocket, according to ACS Publications and The MedChem Digest.

Interface-level evaluation matters because peptide bioactive conformations and interfacial fitness are dictated by the receptor microenvironment, according to ACS Publications.

The closed-loop result matters because it connects computation to experiment: structural hallucination plus REAPS produced candidates that were synthesized, with identity supported by HPLC and mass spectrometry, and functional activity experimentally verified at NK3R, according to ACS Publications.

For U.S. science and technology readers, the practical implication is a openly shared design tool that treats a disease-relevant GPCR pocket as all-atom context from the start, with code, checkpoints, data sets, preprocessing scripts, and a workflow example available for reuse, according to ACS Publications.

The NK3R context underscores clinical relevance because NK3R is already the target of non-hormonal vasomotor-symptom drugs and investigational antibodies, while REAPS explores a different modality: de novo designed peptide binders with experimentally verified functional activity, according to ACS Publications, the Chemicalbook article, and the AbCellera announcement.

Sources referenced