Computational Biology Research

Insulin dynamics, Type 2 Diabetes drug targets, molecular simulations, and systems biology — a research program at the intersection of computational chemistry and metabolic disease.

Molecular Dynamics T2D Drug Discovery GROMACS ChEMBL AlphaFold Cornell BioHPC

Overview

Comprehensive computational investigation of insulin biology and Type 2 Diabetes, leveraging Cornell BioHPC (660 cores, 2.5 PB Lustre), ChEMBL, RCSB PDB, AlphaFold, GEO, and GWAS Catalog.

410 ns

Insulin MD Trajectory

4,865

Drug Compounds (ChEMBL)

PDB Crystal Structures

AlphaFold Predictions

Validated Drug Targets

GEO Expression Datasets

660

BioHPC CPU Cores

Active MD Simulations

Insulin Monomer Molecular Dynamics

410 ns all-atom MD simulation of human insulin monomer (PDB: 1MSO, Lispro variant). GROMACS 2022.1, Amber03 force field, SPC/E water, 300 K, 1 bar. The longest continuous insulin monomer trajectory from our lab, revealing progressive structural instability.

RMSD — Backbone Deviation (nm)

Radius of Gyration (nm)

RMSF — Per-Residue Flexibility (nm)

Solvent Accessible Surface Area (nm²)

Key Findings

Metric	First 100 ns	Last 100 ns	Trend	Interpretation
RMSD	1.300 nm	1.881 nm	↑ Increasing	Progressive structural drift — not equilibrated
Rg	1.845 nm	2.322 nm	↑ Expanding	Monomer unfolding / swelling
SASA	38.54 nm²	38.80 nm²	~ Stable	Surface area relatively constant
RMSF (A-chain)	1.45 – 1.97 nm		High	C-terminal especially disordered
RMSF (B-chain)	0.88 – 1.60 nm		Moderate	Residues 9–16 most rigid (~0.9 nm)

Biological significance: The continuous RMSD increase over 410 ns demonstrates that insulin monomers are intrinsically unstable — consistent with the known biological requirement for zinc-mediated hexamer formation in pancreatic β-cell storage granules. The A-chain's higher flexibility (particularly residues 17–21) suggests it initiates unfolding, while the B-chain core (residues 9–16) provides residual stability.

T2D Drug Target Landscape

ChEMBL v33 analysis of 7 validated Type 2 Diabetes drug targets. 4,865 compounds evaluated for potency (IC50 < 100 nM). DPP-4 emerges as the most druggable target.

Compound Library Size & Potency by Target

Target	ChEMBL ID	Compounds	Potent (<100nM)	Hit Rate	Median IC50	Approved Drugs
DPP-4	CHEMBL2842	909	414	45.5%	49 nM	Sitagliptin, Vildagliptin, Saxagliptin
PPARγ	CHEMBL4093	435	153	35.2%	55.5 nM	Pioglitazone, Rosiglitazone
GLP-1R	CHEMBL284	783	245	31.3%	420 nM	Semaglutide, Tirzepatide
GPR40	CHEMBL3983	559	21	3.8%	170 nM	Fasiglifam (withdrawn)
GCK	CHEMBL235	520	27	5.2%	270 nM	Dorzagliatin (China)
SGLT2	CHEMBL3510	768	7	0.9%	4,422 nM	Empagliflozin, Dapagliflozin
INSR	CHEMBL1981	891	38	4.3%	7,900 nM	Insulin (biologics)

* GLP-1R approved drugs are peptide agonists, not small molecules. Small molecule GLP-1R modulators remain an active research area.

Structural Biology Database

22 curated PDB structures and 20 AlphaFold predictions covering insulin variants, receptor complexes, and T2D drug targets.

Selected PDB Structures

PDB	Protein	Resolution	Method	Relevance
1MSO	Insulin Lispro (T6)	1.0 Å	X-ray	Our MD starting structure, rapid-acting analog
4ZXB	Insulin Receptor Ectodomain	3.3 Å	X-ray	Hexamer production MD (running)
6PXV	Insulin Receptor (full-length, cryo-EM)	3.2 Å	Cryo-EM	DPP-4 production MD (queued)
7KI0	GLP-1R + Semaglutide	2.5 Å	Cryo-EM	Semaglutide binding mode
6X18	GLP-1R + GLP-1 peptide	2.1 Å	X-ray	Endogenous ligand complex
5VEW	GLP-1R + small molecule	2.7 Å	X-ray	Small molecule GPCR modulation
2PRG	PPARγ LBD	2.3 Å	X-ray	Thiazolidinedione binding
7VSI	SGLT2-MAP17	2.95 Å	Cryo-EM	Empagliflozin binding site
5YQZ	Glucagon Receptor	3.0 Å	X-ray	Counter-regulatory target
3I40	Human Insulin	1.85 Å	X-ray	Native insulin reference
1GCN	Glucagon	3.0 Å	X-ray	Counter-regulatory hormone
4CFH	AMPK (active form)	3.24 Å	X-ray	Metformin downstream target

T2D Protein Interaction Network

Large-scale protein–protein interaction network analysis using STRING database. Mapped connectivity of T2D-associated genes to identify hub nodes and functional modules.

13.7M

Interaction Edges

Core Drug Targets

T2D Hub Proteins

T2D Core Protein Targets

Protein	Gene	UniProt	Function	Drug Class
Insulin	INS	P01308	Glucose uptake signaling	Recombinant insulin
Insulin Receptor	INSR	P06213	Tyrosine kinase receptor	Insulin sensitizers
DPP-4	DPP4	P27487	Incretin degradation	Gliptins
GLP-1 Receptor	GLP1R	P43220	Incretin signaling (GPCR)	GLP-1 agonists
SGLT2	SLC5A2	P31639	Renal glucose reabsorption	Gliflozins
PPARγ	PPARG	P37231	Adipogenesis, insulin sensitivity	Thiazolidinediones
Glucokinase	GCK	P35557	Glucose sensing (β-cell)	GK activators

Differential Gene Expression

GEO microarray datasets comparing T2D vs. control pancreatic islet samples. Identifying differentially expressed genes for pathway enrichment and drug target validation.

Dataset	Samples	Platform	Tissue	Status
GSE25724	13 (T2D) / 7 (Ctrl)	Affymetrix HG-U133A	Pancreatic islets	3 significant genes
GSE38642	63	Affymetrix HG-U133 Plus 2	Pancreatic islets	Metadata parsing
GSE41762	77	Affymetrix HG-U133 Plus 2	Pancreatic islets	Metadata parsing
GSE50244	89	Illumina HT-12 v4	Pancreatic islets	Metadata parsing
GSE76894	103	Affymetrix HG-U133 Plus 2	Pancreatic islets	Metadata parsing

Total: 345 samples across 5 datasets. GSE25724 successfully analyzed (T2D vs control split). Remaining datasets require manual metadata curation for group assignment.

AlphaFold Structure Predictions

20 AlphaFold-predicted structures for T2D-related proteins, complementing experimental PDB entries for targets lacking high-resolution crystal structures.

Protein	Gene	Role in T2D	AlphaFold
Insulin	INS	Primary hormone	AF-P01308
Insulin Receptor	INSR	Signal transduction	AF-P06213
GLP-1 Receptor	GLP1R	Incretin pathway	AF-P43220
SGLT2	SLC5A2	Renal glucose transport	AF-P31639
Glucokinase	GCK	Glucose sensor	AF-P35557
PPARγ	PPARG	Insulin sensitization	AF-P37231
DPP-4	DPP4	Incretin degradation	AF-P27487
Glucagon	GCG	Counter-regulation	AF-P01275
TLR4	TLR4	Inflammation in T2D	AF-O00206
AMPK α1	PRKAA1	Metformin target	AF-Q13131

Simulation Campaign

Cornell BioHPC SLURM cluster — active and completed molecular dynamics simulations.

Insulin Monomer 410 ns

Mar 8–23, 2026 · Job 548214 · cbsuecco14 (56 cores) · 212.4 ns/day

PDB 1MSO (Lispro). Amber03/SPC/E. Revealed progressive instability: Rg 1.85 → 2.32 nm. Confirms hexameric storage preference.

Insulin Monomer 100 ns (Pilot)

Mar 8, 2026 · Job 547545 · cbsuecco12 (48 cores) · 277.7 ns/day

Initial exploration. RMSD equilibrated after ~20 ns, then Rg expansion detected — motivated 410 ns extension.

ChEMBL + PDB + GEO + GWAS Batch Analysis

Mar 8, 2026 · Job 547584 · cbsuecco07 (32 cores)

7 drug targets (ChEMBL), 22 PDB structures, 5 GEO datasets (345 samples), GWAS catalog. Full T2D landscape mapping.

Insulin Hexamer 50 ns

Mar 27, 2026 · Job 548385 · cbsuecco01 (16 cores)

PDB 4ZXB (6 chains, 1601 residues). Comparing hexamer vs monomer stability. T = 299.7 K, PE = −1.06×10⁷ kJ/mol.

DPP-4 Full Pipeline (EM→NVT→NPT→50ns)

Mar 27, 2026 · Job 548383 · Queued

PDB 6PXV (6 chains, 1830 residues). Fresh topology rebuild. GROMACS 5.1.2 for large complex handling.

Insulin Lispro 500 ns Extension

Planned — pending GPU reservation (NVIDIA A40)

Extend 410 ns trajectory to determine if RMSD ever reaches a plateau. GPU acceleration expected 10–50× speedup.

Literature Monitoring

Automated daily scan of arXiv, PubMed, bioRxiv, medRxiv, and OpenAlex. Selected high-relevance papers from our continuous monitoring pipeline.

GLP-1R Agonists Activate POMC Neurons ⭐⭐⭐⭐

Merkle lab, Cambridge. Semaglutide causes Ca²⁺ influx via L-type channels in hPSC-derived POMC neurons. Direct human evidence for GLP-1R mechanism. bioRxiv 2024.

Structural Atlas of Human Interactome ⭐⭐⭐⭐

Boltz-2 for de novo binary protein structure prediction at atom scale. Applicable to T2D protein network analysis. bioRxiv 2025.

TZDs vs Dementia in T2D ⭐⭐⭐

Target trial emulation: thiazolidinediones vs other antidiabetics on dementia incidence. PPARγ neuroprotection. medRxiv 2026.

GCN for CVD Risk in T2D ⭐⭐⭐

Interpretable graph convolutional networks for cardiovascular disease risk prediction. Extends network medicine approach. PubMed 2026.

Force Field from Scratch ⭐⭐⭐

Training protein + small molecule force fields de novo. Potential improvement to our MD simulation accuracy. arXiv 2026.

CGM Pre-Diabetes Stratification ⭐⭐⭐

Continuous glucose monitoring in non-diabetics reveals glycemic phenotypes. Pre-diabetes computational modeling opportunity. PubMed 2026.

Methods

Reproducible protocols for all computational analyses.

Molecular Dynamics

Software: GROMACS 2022.1 (CPU), 2024.2 (CUDA 12)
Force field: Amber03 + SPC/E water
Box: Dodecahedron, d = 1.5 nm minimum
Equilibration: EM (steepest descent) → NVT 100 ps (V-rescale 300K) → NPT 100 ps (Parrinello-Rahman 1 bar)
Production: dt = 2 fs, PME electrostatics, LINCS H-bond constraints
Analysis: RMSD, RMSF, Rg, SASA (GROMACS built-in tools)

Cheminformatics

Database: ChEMBL v33
Targets: 7 T2D targets (DPP-4, PPARγ, GLP-1R, SGLT2, INSR, GCK, GPR40)
Filters: IC50, Ki, EC50 assays; potency threshold < 100 nM
Tools: ChEMBL web services API, RDKit, pandas

Structural Biology

PDB: 22 structures (resolution < 4.5 Å)
AlphaFold: 20 predictions (EBI database)
Visualization: PyMOL, ChimeraX
Planned: AutoDock Vina, Schrödinger Glide for virtual screening

Systems Biology

Network: STRING PPI database (13.7M edges)
Expression: GEO microarray (5 datasets, 345 samples)
Genomics: GWAS Catalog (T2D-associated loci)
Structure prediction: AlphaFold DB + ColabFold
Literature: Daily arXiv/PubMed/bioRxiv scan (automated cron)

Compute Infrastructure

Resource	Specs	Use
Cornell BioHPC (ECCO)	12 nodes, 28–112 cores/node, 128–1024 GB RAM, SLURM	MD production runs
BioHPC GPU	2× NVIDIA A40 (48GB), 2× P100, 2× T4	GPU-accelerated MD (pending reservation)
Lustre Storage	2.5 PB (1.2 + 1.3 PB)	Trajectory storage, shared data
Research3 (Johnson)	Kaiko crypto data, WRDS databases	Cross-domain analysis