LAYER / 01 RESOLVEDOrganism
Disease phenotype
Resolve the anatomical and functional signatures that distinguish disease from health.
AI-NATIVE BIOLOGY WORKSPACE
Model life.Decode disease.
MindCell brings virtual cell modeling, disease phenotype intelligence, and specialized biomedical AI into one agent-orchestrated research environment.
- + Agent-routed
- + Model-aware
- + Traceable
LIVE IN-SILICO RESPONSE
06organelles modeled
SCROLL TO DISCOVER
AI-NATIVE DRUG DISCOVERY
From AI disease modeling
to drug discovery.
Build a computable disease model across organism, cell, and molecule—then use the same biological model to screen interventions and identify therapeutic targets.
01Build the disease modelPHENOTYPE · CELL · MECHANISM
LAYER / 01 RESOLVED
LAYER / 02 RESOLVEDCell
Cellular phenotype
Trace tissue-level change to the cell states, morphology, and functions that drive it.
LAYER / 03 RESOLVEDMolecule
Molecular mechanism
Connect cellular dysfunction to actionable pathways, proteins, and molecular events.
02Turn the model into discoveryMODEL · SCREEN · PRIORITIZE
STAGE 01MODEL
AI disease modeling
Build a computable disease model from multimodal phenotype and patient evidence.
+ MULTIMODAL+ CAUSAL
STAGE 02SCREEN
Cellular drug screening
Run in-silico perturbations and prioritize compounds by phenotype rescue, not proxy alone.
+ PERTURB+ RANK
STAGE 03TARGET
Target identification
Surface intervention points linked to mechanism, predicted response, and supporting evidence.
+ MECHANISM+ EVIDENCE
OUTPUTMechanism-linked therapeutic hypotheses
Traceable from phenotype to target.
FOUNDATION MODELS, ONE WORKSPACE
The model hub for living systems.
Run proven biological AI models through one research interface. MindCell routes your question to the right specialist, keeps inputs and outputs together, and makes every result traceable.
MindCellModel HubROUTED BY AI
AlphaFold 3
Complex structure
Boltz
Structure & affinity
OmegaFold
Single-sequence folding
BioEmu
Protein dynamics
OpenFold
Open structure platform
DiffDock
Molecular docking
GNINA
Neural docking
RFdiffusion
Generative design
ProteinMPNN
Sequence design
ProGen2
Protein generation
SaProt
Structure-aware protein LM
ESM
Protein language model
Evo 2
Genomic language model
RiNALMo
RNA language model
ViennaRNA
RNA structure & design
Uni-Mol
Molecular foundation model
MindCell research blog
Ideas, methods, and evidence.
01
How to Map Protein Aggregation Propensity with Aggrescan3D and a PDB Structure
Static Aggrescan3D 1.0.2 analysis of the official RCSB Protein Data Bank structure 2GB1 produced finite scores for all 56 residues of chain A. The minimum score was −3.5629, the maximum was 1.1983, and the arithmetic mean was −1.4135875. The highest-scoring residue was valine A:21 at 1.1983; methion
Read article →02How to Align PDB Protein Structures with PyMOL and Interpret RMSD Correctly
PyMOL 3.1.0 aligned the official RCSB/PDBe 1D3Z ubiquitin NMR structure as the mobile object to the official RCSB 1UBQ X-ray ubiquitin structure as the target. The exact cmd.align return was [0.39735108613967896, 449, 5, 1.2869186401367188, 602, 381.0, 76]: refined RMSD 0.397351 Å across 449 atom pa
Read article →03How to Reconstruct Ancestral DNA Sequences with ancseq and IQ-TREE
The validated workflow selected the IQ-TREE model from the supplied alignment, rooted the analysis with AVR-Mgk5GE162, reconstructed 13 ancestral records, and produced all six required final deliverables. The sorted state table contained 68,807 bytes of explicit state probabilities; intermediate IQ-
Read article →04How to Annotate and Number an Antibody Light Chain with ANARCI and Kabat
ANARCI recognized a human kappa light chain spanning residues 0–106. The retained HMM score was 197.0 with an e-value of 1.6 × 10⁻61, and the assigned germlines were IGKV1-1201 and IGKJ101. Both the Kabat-numbered CSV and the HMM evidence table passed semantic validation. This result passed native e
Read article →05How to Estimate Aqueous Solubility from SMILES with the ESOL Model
The model predicted logS values of −0.0424460527 for ethanol, −1.9325820263 for benzene, and −2.3143366025 for octanol. The expected ordering ethanol benzene octanol was recovered, and the output retained molecular weight, LogP, rotatable bonds, aromatic proportion, SMILES, and units. This result pa
Read article →06How to Infer a Gene Regulatory Network with Arboreto GRNBoost2
GRNBoost2 returned eight ranked transcription-factor–target edges. Both planted relationships, TF1→targetTF1 and TF2→targetTF2, were recovered. The summary retained the expression dimensions, regulator list, seed 777, top edge, recovery flags, and the explicit statement that importance is predictive
Read article →07How to Create and Validate a Minimal BIDS MRI Dataset with PyBIDS
The workflow created a synthetic T1w image, a four-volume resting-state BOLD image, events, sidecars, participant metadata, and dataset metadata. PyBIDS recovered the subject, task, BOLD file, inherited repetition time, and NIfTI shape. The inventory contained 12 rows and the real BIDS validator ret
Read article →08How to Standardize Molecules, Calculate Descriptors, Cluster Fingerprints, and Generate 3D Conformers with Datamol
A validated Datamol workflow standardized six named SMILES records, calculated molecular weight, cLogP, hydrogen-bond donor and acceptor counts, and topological polar surface area, generated 2048-bit radius-2 ECFP fingerprints, calculated a symmetric 6 × 6 distance matrix, assigned every molecule to
Read article →09How to Score a Protein Homodimer Model and Resolve Chain Mapping with DockQ
DockQ 2.1.3 selected the AB:AB model-to-native mapping and returned DockQ 1.000, iRMSD 0.000 Å, LRMSD 0.000 Å, fnat 1.000, F1 1.000, and zero clashes. These exact values are expected because the synthetic model coordinates are identical to the native fixture. This result passed native execution, rea
Read article →10How to Calculate Molecular Electronic Properties with Psi4: Energy, HOMO–LUMO Gap, and Dipole
Electronic-structure calculations turn a molecular geometry, charge, multiplicity, method, and basis set into quantitative predictions about electrons and energy. This worked example uses Psi4 1.11 to perform a restricted Hartree–Fock single-point calculation with the STO-3G basis for neutral single
Read article →11How to Calculate Molecular Electrostatic Potential with Psi4 on a Cartesian Grid
Molecular electrostatic potential, commonly abbreviated ESP, describes the interaction energy per unit positive test charge at positions around a molecule. It combines the attractive contribution of the nuclei with the repulsive contribution of the electron density, producing a spatial field that he
Read article →12How to Compare Protein Sequences with EMBOSS Needle, Water, and seqret
Two protein sequences that differ at only one residue can be compared in several scientifically distinct ways. A global alignment asks how well the complete sequences correspond from end to end, whereas a local alignment asks for the best matching subsequences. Format conversion is a separate operat
Read article →13How to Enumerate an R-Group Compound Library with RDKit, SMILES, and SDF
R-group enumeration generates a virtual compound library by combining a shared molecular scaffold with a defined set of substituents at labeled attachment positions. In the validated example documented here, a phenyl scaffold containing one dummy atom was combined with methyl, amino, and hydroxy sub
Read article →14How to Annotate Antibody Sequences with IgBLAST, AIRR TSV, and Mouse V(D)J Germlines
IgBLAST annotates an immunoglobulin or T-cell receptor sequence by comparing it with curated germline V, D, and J gene databases and by locating the junction, complementarity-determining regions, and framework regions. In the worked mouse immunoglobulin heavy-chain example below, IgBLAST 1.22.0 assi
Read article →15How to Query and Download Public Cancer Imaging DICOM Data with Imaging Data Commons
The National Cancer Institute Imaging Data Commons provides public radiology, pathology, and derived imaging data with searchable metadata and explicit licensing. A reproducible download begins with a narrow metadata query, not a bulk transfer: identify the exact series, inspect its modality, instan
Read article →16How to Perform an ISO 13485 Quality-Management Document Gap Analysis for Medical Devices
Direct answer: The validated document triage found 18 of 26 expected procedure topics, leaving 8 missing topics and a descriptive coverage value of 69.2%. This is a reproducible gap-screening result, not certification, legal advice, or a conformity determination.
Read article →17How to Compare Tandem Mass Spectra with Matchms Modified Cosine Similarity
Three spectra were parsed: two references and one query. Exact greedy cosine returned zero because every query fragment was shifted by 2 Da. Modified cosine accounted for the precursor shift and matched the caffeine reference with score 0.9999999999999999 and five matched peaks; the unrelated refere
Read article →18How to Run a Reproducible GROMACS Energy Minimization on a Tiny Molecular System
GROMACS 2024.5 completed steepest-descent minimization in 27 steps. Potential energy fell from 136.938477 to -0.972452 kJ/mol, and the run reported convergence. Native EDR, TRR, and GRO deliverables were reopened successfully. This result passed native execution, real chat-driven execution, and sema
Read article →19How to Validate an OpenMM CPU Molecular Dynamics Workflow with Energy and Coordinate Checks
OpenMM 8.3.1 selected the CPU platform. Minimization reduced potential energy from 4.5 kJ/mol to 3.8518598887744717e-32 kJ/mol; after 20 one-femtosecond Langevin-middle steps, potential energy was 5.680147519456703e-05 kJ/mol. Two final three-dimensional positions were retained in nanometres. This r
Read article →20How to Triage Compound Libraries with Medchem Rules, Structural Alerts, and Validated SMILES
Five input rows yielded four valid structures and one explicit invalid-SMILES error. Three of four valid structures passed Rule of Five, Veber, PAINS, common-alert, and NIBR checks. Rhodanine failed PAINS and NIBR checks; the long alkane failed Rule of Five, Veber, and common-alert checks. This resu
Read article →RESEARCH COLLABORATION
COSINX
Build models with us.
For research partnerships, scientific collaboration, and MindCell enquiries, contact our team directly.
research@cosinx.com