Technical Deep Dive

METHODOLOGY

A detailed technical reference for how MolForge generates, validates, and ranks drug candidates across the full discovery pipeline.

01

Generative Chemistry: MolForge-Gen

MolForge-Gen is a proprietary 25.4M-parameter Transformer model (GPT-2 architecture) for conditional molecular generation. Unlike off-the-shelf tools, MolForge-Gen natively understands target context, property constraints, and patent boundaries as first-class generation conditions.

  • Architecture: Decoder-only Transformer, 8 layers, 512dim, 8 heads (25.4M params)
  • Pre-trained: 1.6M drug-like SMILES from ChEMBL (Validity 97.5%)
  • Multi-target: Single model generates for TYK2/CDK4/CDK6/TNIK/EGFR via target tokens
  • Property-conditioned: QED, hERG, pIC50, MW ranges as input tokens
  • FTO-aware: Tanimoto distance constraint learned during RL (100% FTO safe)

Model Specifications

Parameters25.4M
ArchitectureGPT-2 Decoder-Only
Pre-train Data1.6M SMILES (ChEMBL)
Pre-train Validity97.5%
Pre-train Uniqueness99.8%
Fine-tune Targets5 (TYK2/CDK4/CDK6/TNIK/EGFR)
TNIK Accuracy99.0%
RL FTO Safe100%
RL Avg QED0.872

02

ADMET Prediction: Dual Model Screening

Every generated compound passes through ADMET-AI (Swanson et al., 2024) for 104-property prediction. Compounds failing any of the five critical thresholds are eliminated before further analysis.

PropertyModelThresholdAction
hERG CardiotoxicityADMET-AI (ChEMBL)> 0.7Eliminate
AMES MutagenicityADMET-AI (ChEMBL)> 0.5Eliminate
HIA (Oral Absorption)ADMET-AI (TDC)< 0.5Eliminate
Molecular WeightRDKit> 500 DaEliminate
PAINS AlertsRDKit SMARTS> 0Eliminate
QEDRDKit< 0.3Eliminate

03

Conformal Prediction: Calibrated Uncertainty

Every pIC50 prediction includes a 90% confidence interval computed via split conformal prediction (Vovk et al., 2005). This is not a Bayesian posterior or a dropout ensemble — it is a distribution-free guarantee that the true value falls within the interval at least 90% of the time on exchangeable test data.

“No other AI drug discovery platform provides distribution-free coverage guarantees on their predictions. This is not a feature — it's a fundamental requirement for responsible computational chemistry.”

Validation Results

Target Coverage90%
Achieved Coverage86.5%
MethodSplit Conformal (Mondrian)
Calibration SetScaffold-split holdout
Interval WidthAdaptive (per-compound)
Confidence LevelsHIGH / MED / WIDE

04

ROBOGATE: Failure Boundary Mapping

ROBOGATE uses adaptive sampling to explore the multi-dimensional property space around a target. Instead of optimizing for a single objective, it maps the Pareto frontier across 7 competing criteria: potency (pIC50), cardiac safety (hERG), mutagenicity (AMES), oral absorption (HIA), drug-likeness (QED), synthesizability (SA Score), and patent novelty (Tanimoto distance).

7

Optimization Axes

500

Pareto Rank-1 (TYK2)

33,755

TNIK Compounds Mapped

05

Boltz-2: Structure Prediction & Docking

Top-ranked compounds are submitted to Boltz-2 (MIT/Recursion, open-source) for protein-ligand co-folding. Compounds achieving ligand_ptm ≥ 0.85 are promoted to Tier 2 (Structure Verified). Boltz-2 provides binding affinity predictions approaching FEP accuracy at 1,000× the speed.

~15s

Per Complex

0.85+

ligand_ptm Threshold

100%

Success Rate (TYK2)

Tier 2

Promotion Level

06

MolForge Score: 7-Axis Compound Ranking

The MolForge Score is a composite metric that ranks compounds across 7 axes using Pareto optimization. Unlike single-objective scoring, it identifies compounds that are not dominated on any axis — the true drug-like sweet spot.

AxisSourceWeightDirection
pIC50 (Potency)QSAR GNNHighMaximize
hERG (Cardiac Safety)ADMET-AIHighMinimize
AMES (Mutagenicity)ADMET-AIHighMinimize
QED (Drug-likeness)RDKitMediumMaximize
SA Score (Synthesizability)RDKitMediumMinimize
HIA (Oral Absorption)ADMET-AIMediumMaximize
Tanimoto (Novelty)RDKit Morgan FPLowMaximize distance

07

Freedom-to-Operate Analysis

Every compound is compared against known patented structures using Morgan fingerprint Tanimoto similarity. For TNIK specifically, all compounds are compared against Insilico Medicine's rentosertib (the only clinical-stage TNIK inhibitor).

TNIK FTO Status:All 33,755 ADMET-validated TNIK compounds have Tanimoto similarity < 0.21 vs rentosertib. Industry standard for structural novelty is Tanimoto < 0.85. MolForge compounds are 4× more structurally distinct than the minimum required threshold.

Model Roadmap

MolForge-Gen v1 → v2

v2 training begins Q2 2026 on RTX 5090 (32GB). Targets ~2x model capacity, 5x pre-train data, and explicit reasoning-style multi-step generation. v1 → v2 metrics will be tracked publicly.

Specv1 (Current)v2 (Q3 2026 Target)Δ
Parameters25.4M~50-100M2-4×
ArchitectureGPT-2 decoder, 8L × 512d × 8hGPT-2 decoder, 12L × 768d × 12hDeeper + wider
Pre-train Data1.6M ChEMBL drug-like~5-10M (ChEMBL + PubChem subset)3-6×
Validity97.5%≥99% target+1.5%
Novelty (vs ChEMBL)96.8%≥98% target+1.2%
Property-ConditioningQED, hERG, pIC50 ranges+ multi-objective (Pareto-aware)Native multi-objective
Targets Fine-Tuned5 (TYK2, CDK4, CDK6, TNIK, EGFR)20+ (Kinome roadmap)
Compute (training)RTX 4090 / ~3 daysRTX 5090 / ~5-7 daysSame generation

Status

Architecture finalized

Data preprocessing in progress

Training start

2026-05

Estimated

First public benchmark

2026-Q3

v2 vs v1 comparison report

Open-Source Tool Stack

MolForge-Gen

Generative Chemistry

Proprietary

Boltz-2

Structure Prediction

MIT

ADMET-AI

ADMET Prediction

MIT

RDKit

Cheminformatics

BSD

AutoDock Vina

Molecular Docking

Apache 2.0

PyTorch

GNN Training

BSD

Open Targets

Target Validation

Apache 2.0

Conformal Pred.

Uncertainty Quantification

Internal