Ziyi Yan

3D structure of circRNA

Circular RNAs lack 5′/3′ termini, so standard linear folding pipelines do not apply directly. This line of work builds a geometric representation on a toroidal manifold — secondary-structure consensus, segmented multi-predictor folding, coarse-grained replica-exchange and metadynamics sampling — followed by all-atom reconstruction and validation.

Structure → immune fingerprints

Turning predicted 3D structures into computable immune-related fingerprints (PKR binding, m6A modification, TLR7 recognition, NLRP3 inflammasome, miRNA sponging) so that structure becomes an intermediate quantity downstream pharmacokinetic and immune-response models can consume directly.

Small-sample drug discovery

Adaptive mixture-of-experts architectures combined with explicit pharmacokinetic dynamics, aimed at screening circular RNA therapeutic candidates under small-sample conditions.

Micro-expression recognition

A bio-inspired dual-channel micro-expression recognition system using a biologically motivated mixture-of-experts structure (BioMoE) to handle fine-grained temporal variation in facial action units.

Torusfold-physics

Physics-based 3D structure prediction for long circular RNA · iGEM 2026, Team JLU-FBH

A physics-based pipeline for long circRNA: secondary-structure consensus, segmented multi-predictor folding, coarse-grained REMD + metadynamics, all-atom reconstruction. Ships a 2,013 nt all-atom model together with the checks that re-derive its reported numbers.

TorusFold-scheme2-rl · torusfold-preview

Toroidal topology · RL completion of long-range pairs

A branch of the work that reconstructs structure on a toroidal topology: long-range pairs are identified by graph distance, and MCTS plus a policy network fill in the pairings that local optima tend to miss.

TorusFold-repro

Reproducible pipeline for circRNA 3D structure prediction

Starting from a single circRNA sequence, produces scoring JSON identical to the server endpoint (physical / immune / circDesign / rsRNASP1) plus 3D structures (coarse-grained and all-atom PDB). Contains everything needed for reproduction and nothing that is not.

confluencia · Confluencia-3.0

Adaptive MoE with Pharmacokinetic Dynamics

An adaptive mixture-of-experts framework for small-sample circRNA drug discovery that models pharmacokinetic dynamics explicitly inside the inference chain, integrated with multi-scale tumour-evolution simulation (including PhysiCell co-simulation).

confluencia-nomad-addon

Screening add-on for offline-first platforms

A Confluencia add-on for Project N.O.M.A.D, bringing circRNA therapeutic candidate screening to offline-first environments.

Censor-MER

Bio-inspired dual-channel micro-expression recognition with BioMoE

Micro-expression recognition with dual-channel feature extraction and a biologically inspired mixture-of-experts structure.

BenchSide

Privacy-preserving social resilience prediction

A system for predicting social resilience among graduate students without centralising raw data: federated graph neural networks model the social-support structure, aligned with safety-constrained reinforcement learning from human feedback.

bioease

R · Bioinformatics analysis toolkit

An end-to-end toolkit covering TCGA data retrieval, differential expression analysis, and survival analysis.

Archiva · IGEM-DryLab-Ambassador

AI ambassador for iGEM-FBH

A conversational agent for iGEM dry-lab outreach: local deployment, retrieval-augmented knowledge-base QA, emotional modelling, and live-stream interaction on Bilibili — built to explain the team's research to a non-specialist audience.

Civis Lucri-Faber

Bio-inspired AI agent

An agent architecture built from biologically motivated mechanisms — curiosity, meta-learning, neuromodulation, epigenetic and metabolic constraints among a dozen others — decomposed into pluggable modules so their effect on long-horizon behaviour can be studied.

manatrix

LLM-Augmented Autonomous Penetration Testing Framework

An LLM-driven password-guessing and automated penetration-testing framework: MAMBA + differential evolution + LLMs for credential generation, with RAG and multi-expert coordination orchestrating the testing workflow.

GMKtec EVO-X3

AMD Ryzen AI Max 395 (Strix Halo) · Radeon 8060S iGPU (gfx1151 / RDNA 3.5)

iGEM 2026 · Team JLU-FBH

Dry Lab · Software Track

Dry-lab member of Jilin University's iGEM 2026 team JLU-FBH: designed and implemented the circular RNA 3D structure prediction pipeline (TorusFold) together with its reproducible toolchain.

Project repository: gitlab.igem.org/2026/software/jlu-fbh/torusfold-hybrid