Researchers at Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, and collaborating institutions have developed Prioritized Integrative Target Optimization (PITO), a computational framework that links human genetic evidence to therapeutic target ranking and herbal candidate discovery. Published in LabMed Discovery , the study uses autoimmune thyroid disease (AITD), hyperthyroidism, and hypothyroidism as exemplars to demonstrate how multimodal genomic integration and network analysis can guide multi-target research in traditional Chinese medicine (TCM).
From genetic associations to ranked targets
The workflow begins with European-ancestry GWAS summary statistics retrieved from the NHGRI-EBI GWAS Catalog using disease ontology terms. Significant lead variants and linkage-disequilibrium proxies are connected to four complementary gene-evidence classes: nearby genes (nGenes); promoter-linked genes supported by promoter capture Hi-C (cGenes); genes supported by expression or protein quantitative trait loci (eGenes); and susceptibility genes inferred through webTWAS (sGenes). These mappings extend variant interpretation beyond physical proximity.
PITO then propagates this evidence through STRING protein–protein interaction networks using random walk with restart (RWR), incorporating peripheral genes connected to genetically supported core genes. Fisher's combined method integrates the resulting predictors into priority ratings on a 0–5 scale. Approximately 15,200 genes were ranked in the overall AITD analysis, creating an ordered starting point for downstream pathway and pharmacology analyses.
Benchmarking against clinical targets
The team constructed a benchmark from ChEMBL AITD drug targets that had reached phase II trials or later. Simulated negatives were sampled from the druggable genome after excluding AITD targets and their interacting neighbors. Evaluation used the area under the sensitivity curve (AUSC) with 1,000 bootstrap iterations for confidence intervals, and the Leading-Gene-for-Target-Discovery (LGTD) metric with 50,000 permutations for enrichment testing. Comparisons with Open Targets Genetics, ClinVar, and a naive druggability baseline supported competitive prioritization performance. The ranking recovered 25 phase-II-or-later AITD targets; 20 of these are also targeted by drugs approved for other indications.
Resolving pathway crosstalk and critical node combinations
Priority ratings were integrated with KEGG-derived gene interactions to identify pathway crosstalk networks, with connections assessed using 100 degree-preserving node permutations. One-sided Fisher exact tests characterized enrichment against KEGG pathways, MSigDB hallmark gene sets, and drug-target annotations.
The overall AITD network contained 30 crosstalk genes. Separate analyses identified 41 genes for hyperthyroidism and 55 for hypothyroidism, comprising 24 shared genes, 17 hyperthyroidism-specific genes, and 31 hypothyroidism-specific genes. Shared genes were enriched in Ras and PI3K/AKT signaling, whereas subtype-specific networks emphasized mTOR/Rap1 in hyperthyroidism and NF-κB/ErbB in hypothyroidism.
Edge-based modularity analysis of the shared network's largest connected component resolved four modules. Stepwise removal of single nodes, pairs, and triples then quantified network fragmentation. Removing EGFR and AKT1 produced a disconnection fraction of 0.667; adding MYC or EP300 increased it to 0.762. These in silico perturbations identify structurally influential combinations for functional testing; they do not establish pharmacological synergy.
Mapping prioritized targets to herbal candidates
Using BATMAN-TCM 2.0, the researchers linked prioritized genes to herbal ingredients and formulas. Cytoscape visualized target–herb–ingredient networks, while degree and betweenness centrality helped identify candidates connected to multiple high-priority targets. Among the top 50 ranked herbs and associated components, 41 had thyroid-disease-related reports in the curated literature. This 82% literature support indicates prior disease relevance rather than clinical efficacy. A supraHex self-organizing map and unsupervised clustering further organized shared and subtype-specific prioritization profiles.
Independent data support and technical implications
In the independent GSE29315 transcriptomic dataset (six Hashimoto's thyroiditis samples and eight controls), seven of the 30 overall AITD crosstalk genes were upregulated at |logFC| > 1 and adjusted P < 0.05, each achieving an individual AUC above 0.7. The small sample size requires further validation. Drug-target mapping also nominated eight hypothyroidism-specific druggable candidates not previously reported in that context: BTK, ERBB2, ERBB3, JAK1, KIT, MET, PDGFRA, and PDGFRB.
PITO provides a traceable route from genetic associations to network-level hypotheses and prioritized TCM experiments. Its technical contribution is connecting disease-supported target selection with the multi-component pharmacology of herbal medicines. This work is entirely computational. Cell and animal studies, diversity replication, compound-target validation, pharmacokinetic and toxicity assessment, and prospective clinical evaluation remain necessary. Optimal combinations and doses have not been established.
The source code is available at https://github.com/hfang-bristol/PITO.
LabMed Discovery
News article
Network-based genetic prioritization identifies targetable hub nodes and herbal compounds bridging hyperthyroidism and hypothyroidism
22-Nov-2026