Siyuan Jiang · BlogPersonal academic website of Siyuan Jiang, an undergraduate student researching computational protein engineering, aptamer design, structure prediction, and peptide virtual screening.https://siyuanj.github.io/blog/2026-10-02T00:00:00+00:00Siyuan Jianghttps://siyuanj.github.io/bioRxiv 2026 | AE-PocketMiner: Using attention to predict cryptic pockets and their allosteric switches at oncehttps://siyuanj.github.io/blogs/2026-10-02-ae_pocketminer_cryptic_pockets.html2026-10-02T00:00:00+00:002026-10-02T00:00:00+00:00AE-PocketMiner adds self-attention to a geometric GNN to predict cryptic pockets and their allosteric coupling from one structure, validated by simulations and thiol labeling on CB1, δ-secretase and VP35.bioRxiv 2025 | NA-MPNN: A nucleic acid inverse folding model that unifies proteins, DNA and RNAhttps://siyuanj.github.io/blogs/2026-10-01-na_mpnn_sequence_design.html2026-10-01T00:00:00+00:002026-10-01T00:00:00+00:00NA-MPNN extends the ProteinMPNN architecture to a single protein-DNA-RNA graph, beating specialized tools at both RNA sequence design and protein-DNA binding specificity prediction.Nature 2026 | GEM: De novo designed GPCR 'exoframe' proteins that control receptor conformation from outside the membranehttps://siyuanj.github.io/blogs/2026-09-30-gpcr_exoframe_modulators.html2026-09-30T00:00:00+00:002026-09-30T00:00:00+00:00De novo designed two-helix transmembrane proteins bind the outer surface of the dopamine D1 receptor and act as anchors, biased, negative or positive allosteric modulators, rescuing loss-of-function mutants.Cell Systems 2025 | ProRNA3D-single: MSA-free, single-sequence protein-RNA complex structure prediction that beats AlphaFold 3https://siyuanj.github.io/blogs/2026-09-29-protein_rna_complex_single_sequence.html2026-09-29T00:00:00+00:002026-09-29T00:00:00+00:00ProRNA3D-single pairs protein and RNA language models through geometric attention to predict protein-RNA complexes without MSAs, outperforming AlphaFold 3 when evolutionary information is scarce.bioRxiv 2026 | T-REX: LLM agents that dynamically orchestrate protein binder designhttps://siyuanj.github.io/blogs/2026-09-28-agentic_binder_campaign.html2026-09-28T00:00:00+00:002026-09-28T00:00:00+00:00T-REX is an LLM-driven campaign controller that switches between rescue, exploration and exploitation during de novo binder design, raising structurally unique hit throughput about 2.4-fold across seven targets.ACS Omega 2025 | AlloBench: A data pipeline and unified benchmark for allosteric site predictionhttps://siyuanj.github.io/blogs/2026-09-27-allobench_allosteric_benchmark.html2026-09-27T00:00:00+00:002026-09-27T00:00:00+00:00AlloBench builds an updatable pipeline producing 2,141 curated allosteric sites and benchmarks 10 prediction tools on a leakage-free test set, finding all fall below 60% accuracy.J. Chem. Inf. Model. 2025 | AlloFusion: Predicting allosteric sites with a protein language model and multi-feature fusionhttps://siyuanj.github.io/blogs/2026-09-26-allofusion_allosteric_site_prediction.html2026-09-26T00:00:00+00:002026-09-26T00:00:00+00:00AlloFusion fuses ProtT5 embeddings, PSSM and biochemical features in a residue-level 1D CNN, predicting allosteric site residues without pocket detection and outperforming pocket-based PASSer methods.bioRxiv 2026 | AutoScreen: Automating target discovery in functional genomics with a multi-agent systemhttps://siyuanj.github.io/blogs/2026-09-25-autoscreen_target_discovery.html2026-09-25T00:00:00+00:002026-09-25T00:00:00+00:00A five-agent pipeline that searches 26 biomedical databases, counters LLM bias toward famous genes, and ranks candidates, recovering about 19% more validated CRISPR screen hits than general-purpose models.Science 2022 | ProteinMPNN: Redefining protein sequence design with deep learninghttps://siyuanj.github.io/blogs/2026-09-24-proteinmpnn_sequence_design.html2026-09-24T00:00:00+00:002026-09-24T00:00:00+00:00ProteinMPNN replaces physics-based energy functions with a message-passing network, raising native sequence recovery from 32.9% to 52.4% and rescuing designs that previously failed experimentally.bioRxiv 2025 | BoltzDesign1: Inverting an all-atom structure prediction model for general biomolecular binder designhttps://siyuanj.github.io/blogs/2026-09-23-boltzdesign1_binder_design.html2026-09-23T00:00:00+00:002026-09-23T00:00:00+00:00BoltzDesign1 freezes Boltz-1 and optimizes the Pairformer distogram instead of backpropagating through diffusion, designing binders for proteins, small molecules, metals, nucleic acids and post-translational modifications.JMB 2025 | Allosteric site prediction: computational methods from sequence to structure and new strategies for drug designhttps://siyuanj.github.io/blogs/2026-09-22-allosteric_site_seq_struct_prediction.html2026-09-22T00:00:00+00:002026-09-22T00:00:00+00:00A review of sequence- and structure-based allosteric site prediction methods and four design cases showing how predicted sites become lead compounds.Nature Computational Science 2024 | RhoDesign: Generative de novo design of RNA aptamers driven by 3D structure predictionhttps://siyuanj.github.io/blogs/2026-09-21-rna_aptamer_generative_design.html2026-09-21T00:00:00+00:002026-09-21T00:00:00+00:00RhoDesign inverse-folds RNA 3D backbones into novel aptamer sequences using RhoFold-predicted training structures, producing fluorogenic Mango-III-like aptamers with hit rates rising from 22% to 75%.eLife 2023 | KeyAlloSite: Using weak coevolutionary signals to find the residues that truly matter in allosteric pocketshttps://siyuanj.github.io/blogs/2026-09-20-coevolution_allosteric_residues.html2026-09-20T00:00:00+00:002026-09-20T00:00:00+00:00KeyAlloSite uses weak, non-contact coevolutionary couplings to rank allosteric pockets and single out the few pocket residues that transmit allosteric signals, supported by asciminib and enzyme engineering cases.bioRxiv 2026 | RFOptimization: Guiding protein design optimization with all-atom structure predictionhttps://siyuanj.github.io/blogs/2026-09-19-rfoptimization_design_optimization.html2026-09-19T00:00:00+00:002026-09-19T00:00:00+00:00A training-free optimizer that rescues borderline de novo designs by alternating RF3 gradient-guided mutation with Boltz prediction and inverse-folding cycles, sharply raising filter-passing rates.bioRxiv 2025 | Ambient Protein Diffusion: Training stronger protein generators on low-quality AlphaFold structureshttps://siyuanj.github.io/blogs/2026-09-18-ambient_proteins_low_quality_structures.html2026-09-18T00:00:00+00:002026-09-18T00:00:00+00:00By letting low-pLDDT AlphaFold structures train only at high diffusion noise levels, a 16.7M-parameter model beats the 200M-parameter Proteina on designability and diversity for long proteins.Protein Science 2026 | DeepSSInter: Predicting inter-protein residue contacts with a structure-aware protein language modelhttps://siyuanj.github.io/blogs/2026-09-17-deepssinter_contact_prediction.html2026-09-17T00:00:00+00:002026-09-17T00:00:00+00:00DeepSSInter combines ESM2, SaProt, and geometric transformers to predict inter-protein residue contacts without MSA, outperforming existing methods in accuracy while being two orders of magnitude faster.J Chem Inf Model 2022 | CorrSite2.0: two dynamic mechanisms of allosteric communication revealed by fast and slow modes, raising prediction accuracy to 97%https://siyuanj.github.io/blogs/2026-09-16-motion_modes_allosteric_prediction.html2026-09-16T00:00:00+00:002026-09-16T00:00:00+00:00Allosteric communication is dominated by fast local modes when the two sites are close and by slow global modes when they are far apart; combining both channels lets CorrSite2.0 identify 97.2% of known allosteric sites.bioRxiv 2026 | BoltzMol-1: Low-budget prospective virtual screening with only dozens of molecules tested per targethttps://siyuanj.github.io/blogs/2026-09-15-boltzmol1_virtual_screening.html2026-09-15T00:00:00+00:002026-09-15T00:00:00+00:00An optimized Boltz-2 pipeline with medicinal chemistry filters and ADMET models identified active molecules or binders for 6 of 10 prospective targets while testing only 28-96 molecules each.bioRxiv 2026 | Odin-Multi: Jointly optimizing one sequence against many targets to control binder specificity and cross-reactivity at the design stagehttps://siyuanj.github.io/blogs/2026-09-14-odin_multi_specific_binder.html2026-09-14T00:00:00+00:002026-09-14T00:00:00+00:00Odin-Multi optimizes a single binder sequence against multiple targets and off-targets inside frozen AlphaFold2, turning cross-reactivity and specificity into design objectives rather than screening outcomes.bioRxiv 2025 | BoltzGen: Unifying sequence and structure through geometric encoding for universal binder designhttps://siyuanj.github.io/blogs/2026-09-13-boltzgen_universal_binder.html2026-09-13T00:00:00+00:002026-09-13T00:00:00+00:00BoltzGen is an all-atom diffusion model that encodes amino acid identity with virtual atoms so sequence and structure are generated jointly, reaching 66% target success on nine novel protein targets.Nature 2024 | AlphaFold 3: One model for predicting nearly every type of biomolecular complexhttps://siyuanj.github.io/blogs/2026-09-12-alphafold3_biomolecular_interactions.html2026-09-12T00:00:00+00:002026-09-12T00:00:00+00:00AlphaFold 3 swaps Evoformer for a Pairformer and adds an atom-level diffusion module, predicting complexes of proteins, nucleic acids, ligands and modified residues more accurately than specialized tools.ICLR 2026 | La-Proteina: All-atom de novo protein generation with partially latent flow matchinghttps://siyuanj.github.io/blogs/2026-09-11-la_proteina_atomistic_generation.html2026-09-11T00:00:00+00:002026-09-11T00:00:00+00:00La-Proteina models Cα coordinates explicitly and compresses sequence and side chains into 8-dimensional per-residue latents, reaching 68.4% all-atom co-designability and generating proteins up to 800 residues.Briefings in Bioinformatics 2026 | Protein–aptamer prediction benchmark: AI models assemble stable complexes but cannot tell true binders from false oneshttps://siyuanj.github.io/blogs/2026-09-10-aptamer_complex_prediction_benchmark.html2026-09-10T00:00:00+00:002026-09-10T00:00:00+00:00Benchmarking AlphaFold3, Chai-1, Boltz-2 and RF2NA on 11 protein–aptamer complexes shows that mismatched and shuffled aptamers also yield stable, favorable complexes, so the models miss binding specificity.Nat Methods 2026 | AF2BIND: Predicting small-molecule binding sites from AlphaFold2's internal representationshttps://siyuanj.github.io/blogs/2026-09-08-af2bind_binding_sites.html2026-09-08T00:00:00+00:002026-09-08T00:00:00+00:00AF2BIND feeds AlphaFold2 a target protein plus 20 amino-acid baits and trains a logistic regression on the pair representation to predict small-molecule binding residues de novo.J Mol Biol 2022 | Dissecting slow modes reveals multilayered information sources of allosteric communication in proteinshttps://siyuanj.github.io/blogs/2026-09-07-slow_modes_allosteric_sites.html2026-09-07T00:00:00+00:002026-09-07T00:00:00+00:00Splitting GNM slow modes into subsets and computing transfer entropy identifies global information sources that match known active and allosteric sites in 23 proteins, revealing layered, bidirectional allosteric pathways.Nat Methods 2024 | RhoFold+: Predicting single-chain RNA 3D structures with an RNA language modelhttps://siyuanj.github.io/blogs/2026-09-06-rhofold_rna_3d_structure.html2026-09-06T00:00:00+00:002026-09-06T00:00:00+00:00RhoFold+ fuses an RNA language model pretrained on ~23.7 million sequences with MSA information to predict single-chain RNA 3D structures, outperforming existing methods including AlphaFold3.ICML 2026 | SwitchCraft: Designing multistate protein switches with programmable constraintshttps://siyuanj.github.io/blogs/2026-09-05-switchcraft_state_switching.html2026-09-05T00:00:00+00:002026-09-05T00:00:00+00:00SwitchCraft frames multistate protein design as sequence optimization through Boltz-1 under composable constraints, producing allosteric switches, ligand discriminators and de novo fluorescent biosensors in silico.bioRxiv 2025 | RFdiffusion3: An all-atom diffusion model unifying protein, DNA, small-molecule and enzyme designhttps://siyuanj.github.io/blogs/2026-09-05-rfdiffusion3_allatom_interactions.html2026-09-05T00:00:00+00:002026-09-05T00:00:00+00:00RFdiffusion3 diffuses every polymer atom explicitly, unifying binder, protein-DNA, small-molecule and enzyme active-site design in a 168M-parameter model that runs 10x faster than RFD2.Nat Methods 2026 | RFdiffusion2: from residues to atoms, a paradigm shift in enzyme active site designhttps://siyuanj.github.io/blogs/2026-09-05-rfdiffusion2_enzyme_scaffolding.html2026-09-05T00:00:00+00:002026-09-05T00:00:00+00:00RFdiffusion2 diffuses directly on atomic coordinates, scaffolds all 41 active sites of the AME benchmark, and yields five experimentally active designed enzymes from minimal chemical constraints.Nature 2023 | RFdiffusion: From noise to protein, diffusion models redefine protein designhttps://siyuanj.github.io/blogs/2026-09-05-rfdiffusion1_protein_design.html2026-09-05T00:00:00+00:002026-09-05T00:00:00+00:00RFdiffusion turns a structure prediction network into a generative one by denoising fine-tuning, solving monomer, symmetric assembly, motif scaffolding and binder design with roughly 100-fold higher experimental success than Rosetta.