From Target to Preclinical Candidate
in Weeks, Not Years

Multi-omics analysis integrating genomics, proteomics, and transcriptomics identifies disease-driver genes beyond literature bias. Targets are scored using a proprietary quantitative Target Prioritisation Matrix (TPM). Protein structures are retrieved from PDB or predicted by AI for novel targets, then subjected to conformational ensemble generation via MD. Binding pockets are identified by a multi-tool consensus approach including geometry-based detection, energy-based scoring, Gaussian density mapping, and MD-driven cryptic pocket detection on the conformational ensemble — capturing allosteric and transiently-open pockets that static methods miss.

All compounds pass a tiered filter cascade before docking: physicochemical filters (Ro5, Veber), PAINS removal, reactive group elimination, structural toxicity alerts, and indication-specific ADMET gating. Remaining compounds proceed to consensus docking across multiple industry-standard platforms, covalent docking for Cys/Ser/Lys-targeting warheads, induced-fit docking for receptor plasticity, pharmacophore-based ligand screening, and deep learning rescoring post-docking for improved rank correlation.

Top-ranked poses are clustered for scaffold diversity, then subjected to short MD simulations (20–50 ns) for pose stability and binding mode validation. Solvation thermodynamics analysis captures water displacement contributions — a major affinity driver invisible to docking and standard MM-GBSA. MM-PBSA rescoring is applied to the shortlisted top hits, followed by AI-enhanced proprietary pose refinement. Gate criterion: ≥20 hits with stable MD pose and ΔG < −7 kcal/mol.

Confirmed hits undergo rigorous alchemical free energy perturbation (ABFE and RBFE, <1 kcal/mol uncertainty) to guide quantitative SAR. Scaffold hopping, fragment elaboration, matched molecular pair analysis, and bioisosteric replacements expand chemical space while maintaining binding mode. Every proposed analogue passes a two-tier synthesis feasibility assessment: AI-driven retrosynthetic route generation with yield prediction, commercial building block availability check, and IP freedom-to-operate screening. Red IP flags or synthesis failures are replaced before advancing.

Comprehensive in silico ADMET profiling covering solubility, Caco-2 permeability, plasma protein binding, BBB penetration, P-gp, oral bioavailability, Vd, and t½. DMPK profiling covers all major CYP isoforms, metabolic soft-spot prediction, intrinsic clearance, reactive metabolite flags, and Phase II metabolism. Safety flags include hERG, Ames, DILI, QT prolongation, and skin sensitisation. Full PBPK modelling projects human systemic and tissue exposure across dose levels. All parameters integrated into a composite MPO scorecard with Pareto-front visualisation aligned to the agreed Target Product Profile.

Targets are assessed by a proprietary Target Prioritisation Matrix integrating multi-omics, literature databases, and quantitative genetic/functional evidence. Experimental PDB structures are retrieved and quality-filtered; AI-predicted structures are generated for novel or under-crystallised targets. Conformational ensembles capture flexibility, loop dynamics, and transient pocket opening. Epitope mapping employs B-cell accessibility prediction, T-cell MHC-I/II binding affinity, PPI hotspot alanine-scanning, buried surface area analysis, and cross-species conservation overlay — producing a binding interface atlas with hotspot heatmaps and selectivity risk flags.

The core of the platform: proprietary diffusion-based generative AI designs backbone scaffolds conditioned directly on the target binding interface — no template or known binder required. 10,000–100,000 backbone proposals are generated per target site. A proprietary inverse-folding neural network then designs amino acid sequences predicted to fold stably into the target geometry, with multi-sample diversity at varying sampling temperatures. Antibody and nanobody design uses specialist immunoglobulin AI covering VHH, VH/VL paired design, CDR-H3 loop generation, bispecific formats, Fc engineering, and humanisation. All designed sequences undergo self-consistency validation: TM-score ≥0.85 and interface confidence ≥80 required to advance.

A tiered filter cascade removes low-quality, redundant, or developability-compromised candidates: sequence clustering at 80% identity, structure confidence thresholds, pI filtering (5.5–8.5), hydrophobicity and aggregation propensity prediction, thermal stability (Tm >60°C), cysteine mapping, and chemical degradation site scanning (deamidation, oxidation, isomerisation). Filtered candidates proceed to protein–protein docking with rigid-body and flexible ensemble protocols, interface energy decomposition, buried surface area thresholds (≥600–1,600 Ų by modality), shape complementarity (Sc ≥0.60), and hotspot contact coverage. Gate criterion: ≥20 candidates passing all pre-filter criteria.

All-atom MD simulation (100–500 ns, GPU-accelerated) validates binding mode stability, interface contact persistence (>70% trajectory duration), and thermal stability. Solvation thermodynamics analysis is integrated. For leads progressing to candidate status, alchemical FEP provides ABFE (ΔG uncertainty ≤1 kcal/mol) and RBFE (ΔΔG uncertainty ≤0.5 kcal/mol) for quantitative structure–activity guidance. Iterative sequence optimisation via proprietary inverse-folding redesign, computational saturation mutagenesis, and directed evolution simulation achieves 2–10 fold improvement in predicted ΔG per optimisation round. Proteome-wide homology scan and paralogue cross-docking ensure selectivity margins.

Comprehensive in silico biophysical developability panel: solubility, colloidal stability, SAP aggregation scoring, thermal stability (Tm >65°C threshold), high-concentration viscosity, chemical stability (deamidation/oxidation/isomerisation), and polyreactivity prediction. Multi-layer immunogenicity assessment covers T-cell epitope mapping across 15+ HLA-DR/DP/DQ alleles (>90% global population coverage), B-cell epitope prediction, antibody humanness scoring, germline deviation, and in silico deimmunisation where required. CMC assessment covers expression system compatibility (E. coli/Pichia/CHO/HEK293), signal peptide optimisation, disulfide mapping, glycosylation prediction, proteolytic stability, and formulation compatibility.

PK/PD modelling covers FcRn-mediated half-life (IgG/Fc-fusions), renal clearance prediction, serum stability, tissue penetration (superior for nanobodies vs. full IgG), target-mediated drug disposition, and full PBPK simulation for human dose projection. Therapeutic format engineering designs half-life extension (Fc fusion, albumin binding, PEGylation sites), bispecific antibodies, bivalent VHH tandems, ADC conjugation sites, pH-sensitive antigen release, and cell-penetrating fusions for intracellular targets. All outputs integrated into a composite MPO scorecard with Pareto-front visualisation across binding affinity, developability, immunogenicity, and PK suitability.