Computational De Novo Design of Miniprotein Candidate Neutralizers Targeting the IL-23 Cytokine Subunit p19

Raghav Paladugu; Fardina Fathmiul Alam

AI summary · arXiv: 2610.32779v1 · q-bio.BM

Computational De Novo Design of Miniprotein Candidate Neutralizers Targeting the IL-23 Cytokine Subunit p19

Raghav Paladugu, Fardina Fathmiul Alam

Original paper Submitted to arXiv on 23 Oct. 2026 Accepted as a short paper at the 2026 IEEE International Conference on Bioinformatics and Biomedicine (BIBM 2026) License: CC BY 4.0

How to read this summary? Illustrated guide

The essentials

The study computationally proposes IL-23 p19-binding miniproteins, finding interface-optimized design far outperformed a conventional pipeline, but experimental validation remains necessary.

Question

Can de novo miniproteins be computationally designed to occlude the IL-23 receptor-binding face of cytokine subunit p19? How do conventional and interface-optimized design pipelines compare when evaluated using a shared interface-focused score?

Method

The authors used the p19 structure from PDB 3DUH, targeting six IL-23R-face hotspots, and generated 972 valid conventional RFdiffusion/ProteinMPNN designs and 50 accepted BindCraft designs. They rescored designs with AlphaFold-initial-guess predictions and ipSAE, assessed conventional interfaces with AlphaFold-independent geometry, characterized five leads using structural and Rosetta metrics, and screened those leads against p40.

Findings

Result Single-sequence ColabFold substantially under-rated binder pLDDT compared with AlphaFold-initial-guess for 50 matched designs.
See the passage in the paper
on 50 matched designs mean binder pLDDT was 30.1 versus 86.3, a mean under-rating of 56 points
Result None of the 972 conventional designs reached the study's reliable-interface ipSAE threshold.
See the passage in the paper
the best binder–p19 ipSAE was 0.297, the median 0.005, and 0/972 reached 0.60 (none reached even 0.40)
Result BindCraft produced 42 designs above ipSAE 0.60, including 38 at or above 0.70.
See the passage in the paper
Of 50 Accepted designs (42 p19-only, 8 full heterodimer), 42 cleared the 0.60 threshold, 38 of them in the $\geq$0.70 confident band
Result The five characterized leads are all-helical and collectively contact all six targeted hotspots.
See the passage in the paper
All five are compact all-$\alpha$-helical bundles of 72–104 residues.
Authors’ interpretation The authors interpret the lead designs as likely to block receptor binding by sterically occluding the IL-23R footprint, but this mechanism has not been experimentally tested.
See the passage in the paper
The neutralizer designation rests on where and how much the binders bury, not on a functional assay: the hotspots were selected as IL-23R contact residues, so a binder burying 2204–3018 Å2 over that face (Table III) occludes the receptor footprint sterically, making steric competition the most parsimonious predicted mechanism until a competition or STAT3 assay is performed.

Limitations

The proposed binders and neutralization mechanism are computational predictions and require experimental validation.
See the passage in the paper
All results are computational: ipSAE is a predictor, not a guarantee, and designed binders can still fail experimentally.
The study did not measure binding affinity.
See the passage in the paper
The target is a single, partially modeled crystal conformation (p19 has unmodeled loops at 30–46, 95–99, 126–136), and we did not measure affinity.
The p40 counter-screen did not establish that the predicted p19-bound poses are sterically compatible with p40 in intact IL-23.
See the passage in the paper
Finally, our selectivity test asks only whether a binder forms a spurious interface with isolated p40; steric compatibility with p40 inside the intact covalently linked cytokine is a separate assembly-level question we did not evaluate.

Why it matters

The authors present what they describe as the first computationally proposed de novo miniprotein candidates targeting the IL-23 cytokine's receptor-binding face, extending miniprotein design beyond receptor-directed approaches. Their comparison also highlights interface-optimized design and interface-aware scoring as useful for generating and assessing candidate binders, while positioning the leads as starting points for experimental testing.

Written by AI from the paper. Passages are checked word for word against the text we analysed; the rest may contain errors.

AI-generated keywords: IL-23 p19 de novo miniproteins interface optimization computational protein design

Read further

Comprehensive Summary
Layman's Summary
Blog article
Original abstract

This paper investigates whether de novo-designed miniproteins can be computationally developed to block the IL-23 receptor-binding surface of p19, the subunit unique to IL-23. Previous miniprotein work in this inflammatory pathway had targeted the receptor or IL-17 rather than IL-23 itself. The authors also examine how design pipelines and interface-confidence scores influence assessments of predicted binders. They stress that their candidates are computational designs, not experimentally validated neutralizers. The target structure was taken from PDB 3DUH. The authors selected six residues on the p19 IL-23R-binding surface as design hotspots: C51, C80, C81, C89, C137, and C140. They compared two design approaches using the same interface metric, ipSAE, calculated from AlphaFold-initial-guess predictions and predicted aligned-error matrices. Scores of at least 0.60 were treated as indicating a reliable predicted interface, and scores of at least 0.70 as indicating a confident one. The authors note, however, that these thresholds have not been calibrated specifically for de novo binders. The first approach used a conventional pipeline: RFdiffusion generated hotspot-conditioned backbones, ProteinMPNN assigned sequences, and ColabFold initially folded the designs. Of 1,000 generated backbones, 972 passed filtering and sequence threading. None reached an ipSAE score of 0.60; the highest score was 0.297, and the median was 0.005. The authors found that the apparent low foldability of some designs was partly a scoring artifact. For 50 matched designs, single-sequence ColabFold produced a mean binder pLDDT of 30.1, compared with 86.3 using pose-seeded AlphaFold-initial-guess—a difference of 56 points on average. They therefore interpret the conventional pipeline’s poor results as a failure to form strong predicted interfaces, not as evidence that its binders cannot fold. The second approach, BindCraft, jointly optimized the binder and interface through AlphaFold2-multimer-based hallucination, followed by sequence redesign and filtering. It produced 50 accepted designs. Using the shared ipSAE evaluation, 42 scored at least 0.60, 38 scored at least 0.70, and the top design reached 0.804. The authors identified 23 distinct backbones among the reliable p19-only designs. The contrast persisted across thresholds from 0.50 to 0.80: the conventional pipeline produced no designs at any cutoff, whereas BindCraft produced 45, 42, 38, and 2, respectively. In an additional geometry-based assessment, none of the 20 top-ranked conventional designs matched even the weakest of five selected leads in both buried surface area and interface hydrogen bonds. The five leads are 72–104 residues long and are predicted to form compact, predominantly alpha-helical bundles. Collectively, they contact all six targeted hotspots, although no individual lead covers all six. Two were designed in the presence of the intact p40/p19 heterodimer. Lead l92 contacts five hotspots in that context and is highlighted as particularly relevant to intact IL-23. Rosetta interface estimates were favorable for all five leads, but the metrics differed on which was best: l94 had the highest ipSAE, whereas l72 had the most favorable estimated interface energy. In a counter-screen against isolated p40, all five scored an ipSAE of 0.00. This supports the absence of a detectable interface with p40 alone, but does not establish that each design is compatible with p40 in intact IL-23.

Designing Miniproteins to Target IL-23

Interleukin-23 (IL-23) helps drive inflammatory immune responses and is an important target in research on inflammatory diseases. Its p19 subunit is unique to IL-23 and contributes to the cytokine’s interaction with its receptor. In “Computational De Novo Design of Miniprotein Candidate Neutralizers Targeting the IL-23 Cytokine Subunit p19,” Raghav Paladugu and Fardina Fathmiul Alam investigate whether small, computationally designed proteins could bind the p19 surface involved in receptor recognition.

Earlier miniprotein efforts in this pathway had focused on the receptor or on IL-17, rather than directly targeting IL-23. This study explores that less-traveled route while also examining an important methodological question: how much do design pipelines and interface-scoring methods shape the apparent quality of proposed binders?

Two Routes to Candidate Binders

The researchers used the IL-23 structure in PDB 3DUH and selected six p19 residues as design hotspots: C51, C80, C81, C89, C137, and C140. These sites lie on the surface involved in binding the IL-23 receptor. The team compared two computational approaches, evaluating their results with the same interface-confidence measure, ipSAE.

For the conventional design pipeline, RFdiffusion generated backbones guided by the selected hotspots. ProteinMPNN then assigned sequences, and ColabFold was used for initial folding predictions. Of 1,000 generated backbones, 972 passed filtering and sequence-threading steps. Yet none reached an ipSAE score of 0.60. The top score was 0.297, and the median was just 0.005.

Those low scores do not necessarily mean the designs could not fold. The authors found that the assessment method made a striking difference: for 50 matched designs, single-sequence ColabFold gave an average binder pLDDT of 30.1, while pose-seeded AlphaFold-initial-guess predictions averaged 86.3. That 56-point gap suggests that some designs appeared poorly folded under one setup but much more structurally confident under another. The authors therefore interpret the conventional pipeline’s results primarily as weak predicted interfaces, rather than proof that its candidates are incapable of folding.

BindCraft Produces Stronger Predicted Interfaces

The second approach, BindCraft, jointly optimized the binder and its interface using AlphaFold2-multimer-based hallucination, followed by sequence redesign and filtering. It yielded 50 accepted designs. Evaluated with the shared ipSAE approach, 42 reached at least 0.60, a threshold treated as indicating a reliable predicted interface. Thirty-eight scored at least 0.70, the study’s mark for a confident interface, and the best reached 0.804.

The difference between the pipelines remained clear across several cutoffs. At thresholds of 0.50, 0.60, 0.70, and 0.80, the conventional pipeline produced no designs meeting the threshold. BindCraft produced 45, 42, 38, and two, respectively. The authors also identified 23 distinct backbones among the reliable p19-only designs, suggesting the higher scores were not limited to a single structural solution.

These cutoffs should be interpreted cautiously. The authors note that the 0.60 and 0.70 thresholds have not been calibrated specifically for de novo-designed binders. They are useful for comparing predicted interfaces in this study, but do not establish that a candidate will bind in a laboratory or neutralize IL-23.

Five Leads and the Intact Cytokine Context

The study highlights five candidate leads, each 72 to 104 residues long. Their predicted structures are compact bundles composed mainly of alpha helices. Collectively, the five designs contact all six selected hotspots, although no single lead covers every hotspot.

Two leads were designed with the intact p40/p19 heterodimer present. This matters because p19 exists as part of IL-23, not only as an isolated protein. Lead l92 contacts five hotspots in that context and is singled out as particularly relevant to the intact cytokine. Rosetta interface estimates were favorable for all five leads, but different metrics favored different candidates: l94 had the highest ipSAE score, while l72 had the most favorable estimated interface energy. The results illustrate why a single score may not capture every aspect of a predicted design.

In a counter-screen against isolated p40, all five leads scored an ipSAE of 0.00. This supports the absence of a detectable predicted interface with p40 alone. It does not, however, prove that each design can bind p19 while remaining compatible with p40 in the complete IL-23 complex. The distinction is important: a negative result against one isolated subunit cannot settle how a candidate behaves in the assembled cytokine.

What the Findings Do—and Don’t—Show

The study’s central result is a computational comparison: in this target and under the authors’ evaluation framework, BindCraft generated candidates with much stronger predicted interfaces than the conventional RFdiffusion-based pipeline. The work also shows that folding confidence can vary substantially with the prediction setup, making it risky to treat a single model or score as a definitive verdict.

Most importantly, these miniproteins remain computational designs, not experimentally validated neutralizers. Their predicted contacts and interface scores make them candidates for further study, not proven treatments or confirmed blockers of IL-23 signaling. Experimental tests will be needed to determine whether they bind p19, work in the intact cytokine, and actually prevent receptor engagement. The paper offers a promising design strategy while keeping that gap between prediction and biological validation firmly in view.

Created on 30 Oct. 2026

Was this synthesis useful?

Score: 0

Ask questions about this paper to our AI assistant

How to chat with this paper? Illustrated guide

Compare with the PDF

You can also chat with multiple papers at once here.

AI assistant instructions?

Tools for this paper

Explain it differently 3 free rewrites left today

Save to a collection Group papers in collections to compare them, chat with them and export them. Create a free account Log in

Export BibTeX · RIS (Zotero, Mendeley, EndNote) · Markdown (Notion, Obsidian)

Advanced tools Premium

For researchers who follow a field closely: citation alerts, journal-club slides, version comparison and Zotero export. Discover Premium

Zotero

Citation alerts Citation alerts guide

Who works on it? The people and institutions who publish on its topic

Weekly letter Presented in "Biology and medicine this week" of 26 October 2026. Biology and medicine this week

Journal club Premium+ Slides to present it

Versions What changed between versions

Talk about it in public Premium+ Press kit: release, questions, hype check Press kit guide

Papers around this one

See them on a map

Ranked by how close their title and abstract are to this paper’s. Select some to read them together.

Closest papers

Very close
Design of a high-affinity IL-21 decoy receptor as a novel therapeutic strategy for autoimmune disorders.
Sara Vazifeshenas, Z. Hajihassan, Seyed Shahriar Arab et al. · 2026 · Journal of Biomolecular Structure and Dynamics · Not summarized yet

Abstract
Elevated levels of the pro-inflammatory cytokine Interleukin 21 (IL-21) contribute to the progression of autoimmune and inflammatory diseases by binding to its receptor (IL-21R) and activating detrimental signaling pathways in immune cells. Consequently, we developed a recombinant decoy receptor protein designed to competitively inhibit IL-21 binding to its receptor. Using the extracellular domain of IL-21R as a scaffold, five key binding-site residues (33, 38, 70, 94, and 130) were mutated with OSPREY software. Computational analyses with GROMACS selected the Q33S, E38D, M70T, L94N, and M130S mutant for its superior structural stability and high binding affinity to IL-21, with binding energ…
Close
Computational Design and Evaluation of De Novo Protein Minibinders Targeting Specific Sites of STAT3
Emily R. Sragovicz, A. Tobias · 2026 · bioRxiv · Not summarized yet

Abstract
STAT3 is a transcription factor involved in the progression of many types of cancer, which makes it an attractive therapeutic target. Small-molecule STAT3 inhibitors have experienced a high failure rate in clinical trials, indicating alternative approaches may be warranted. To address these limitations, we used BindCraft, a modern AI-driven protein design tool, to create miniprotein binder candidates against several sites of STAT3. We selected seven target hotspots on the SH2, linker, DNA-binding and coiled-coil domains of STAT3. Of 139 designs that passed BindCraft’s filters, 11 high-confidence binders emerged from five of the seven hotspots. These candidate binders were evaluated with Alph…
Close
Engineering a hyperstable synthetic miniprotein for B7-H3 targeting
Hannah K. Windsor, Nicholas C. Ulen, Claudia Beaudry et al. · 2026 · Protein engineering, design & selection : PEDS · Not summarized yet

Abstract
Abstract Small protein ligands, such as miniproteins, have the potential to provide the efficient physiological transport and rapid clearance required for radioligand therapy. Yet, binders require single-digit nanomolar affinity for potent targeting and high stability for radiolabeling and physiological integrity. High-affinity and stable single-domain miniproteins are also modular candidates for multispecific applications in cell engagers and avid constructs. Previous work demonstrated computationally designed 40-amino acid β⍺ββ miniproteins as effective scaffolds for ligand discovery. Herein, to yield the desired single-digit nanomolar affinity for B7-H3, an important immune regulator and …
Close
Agent-guided epitope excision enables economical de novo binder design
Jun W. Kim, Artiom Myaskovsky, O. Donosa et al. · 2026 · Not summarized yet

Abstract
De novo protein design campaigns remain bottlenecked by the inference compute time required to generate and rank candidates, limiting wider functional exploration. Here, we introduce and evaluate an epitope excision approach that truncates targets to local interface windows to accelerate both backbone diffusion and structure prediction. Across 15 targets and 3,666 designs, excised windows closely reproduced full-target binding interfaces (median C\alpha RMSD of 0.37 \AA) and made it practical to design binders against the 719-residue extracellular domain of fibroblast activation protein alpha (FAP) by cutting RFdiffusion and AlphaFold3 GPU time by ~ 88% in one of the epitopes. Pairing this w…
Close
In Silico Structural Dynamics and Downstream Bioprocess of Recombinant Human SUMO-IL-33
K. Nguyen, T. Mai, Thua-Phong Lam et al. · 2026 · Biomolecules · Not summarized yet

Abstract
Interleukin-33 (IL-33) is a therapeutic target in inflammatory diseases, yet inhibitor screening is hindered by high commercial protein costs. While expressing IL-33 as a SUMO fusion protein overcomes yield limitations in Escherichia coli, whether this intact chimera can directly serve as a screening surrogate remains unresolved. Here, we combined in silico simulations with empirical bioprocess optimization to establish a high-yield pipeline for native human IL-33 production. AI-based structure predictions (AlphaFold and RoseTTAFold) revealed distinct inter-domain dynamics mediated by a flexible linker. Molecular dynamics simulations of the compact RoseTTAFold model suggested potential steri…
Close
Mirror-Score: Calibrated, Inference-only Scoring Exposes the Limits of Sequence-compatibility Ranking in D-peptide Design
Jian-Dang Li · 2026 · Not summarized yet

Abstract
D-peptides combine protease resistance with high target specificity, but computational design of D-peptide binders remains immature. Mirror-Peptidizer introduced an in silico mirror-image screening pipeline using target reflection, backbone generation, and ProteinMPNN sequence design, but its raw ProteinMPNN negative log-likelihood (NLL) ranking was not validated against measured affinities, and only 4 of 9 tested MDM2 designs bound detectably. We introduce Mirror-Score, a calibrated, inference-only scoring framework for heterochiral D-peptide/L-protein complexes, and a public benchmark of 31 crystal complexes across four target families, including 18 with literature-verified affinities. Raw…
Close
Discovery of Small Peptides as Dual Inhibitors of IL-4 and IL-13 Signaling
Cha-Se Mei, P. Atawuchugi, Tawshik W. Iqbal et al. · 2026 · ACS Medicinal Chemistry Letters · Not summarized yet

Abstract
Interleukin-4 (IL-4) and interleukin-13 (IL-13) signaling are central drivers of type 2 inflammation. While monoclonal antibodies targeting this pathway are clinically effective, their use is limited by high production cost, poor tissue penetration, increased risk of adverse effects, and parenteral administration requirement. Peptides offer advantages including structural tunability, favorable safety profiles, and improved tissue accessibility. Here, we employed an alternating phage display biopanning strategy to identify small peptides capable of dual modulation of IL-4 and IL-13 signaling. Sequential selection against IL-4 receptor α (IL-4Rα) and IL-13Rα1 enriched ligands with intrinsic du…
Close
Topology-enabled long-acting interleukin-2 for advanced low-dose interleukin-2 therapy of autoimmune diseases
Jing Fang, Rui-Ling Feng, Xi-Bao Tian et al. · 2026 · The Innovation Life · Not summarized yet

Abstract
The therapeutic efficacy of protein biologics is often hampered by their rapid in vivo degradation and off-target effects. Interleukin-2 (IL-2), a cytokine promising for treating autoimmune diseases via regulatory T cell (Treg) expansion, exemplifies this challenge due to its short half-life and pleiotropic activity. Herein, we report a molecular-scale topology engineering strategy to overcome these limitations. By using split intein mediated ligation and genetically encoded entangling motif, p53dim (X), we constructed cyclic (c-IL-2) and catenated (cat-IL-2) variants of IL-2 via direct expression in E. coli. These topological proteins preserve the native conformation and receptor binding ca…

Go further

How are these papers found?

We compare the title and abstract of this paper with those of every paper summarized on this site, and with the papers Semantic Scholar recommends, those it cites and those citing it (for a much-cited paper, its most recent citations and, from OpenAlex, the most cited ones). Among the prior work and the follow-ups, the papers that use each other in depth, a method it uses and the much-cited follow-ups come a little higher. The levels (very close, close, related) come from the similarity of the texts, not from a judgment of quality; * marks a paper without an abstract, compared by its title. The sentence under a paper is written by an AI from the two abstracts: check it in the papers. A paper marked not relevant by three readers leaves the list.

Updated on 31 October 2026.

Disclaimer: The AI-based summarization tool and virtual assistant provided on this website may not always provide accurate and complete summaries or responses. We encourage you to carefully review and evaluate the generated content to ensure its quality and relevance to your needs.

Computational De Novo Design of Miniprotein Candidate Neutralizers Targeting the IL-23 Cytokine Subunit p19

The essentials

Question

Method

Findings

Limitations

Why it matters

Read further

Definitions

Designing Miniproteins to Target IL-23

Two Routes to Candidate Binders

BindCraft Produces Stronger Predicted Interfaces

Five Leads and the Intact Cytokine Context

What the Findings Do—and Don’t—Show

Ask questions about this paper to our AI assistant

Tools for this paper

Advanced tools Premium

Papers around this one

Closest papers

Go further

Cite this paper (BibTeX)