VIDraft/pxr-challenge-method

PXR Activity Prediction — Method Report

Team: VIDraft
Website: https://www.vidraft.net
Contact: arxivgpt@gmail.com
Challenge: OpenADMET PXR Blind Challenge — Activity Prediction Track
Submitted: 2026-06-11

---

Abstract

We present an XGBoost-based ensemble approach for predicting PXR (Pregnane X Receptor) agonist activity (pEC50) using molecular fingerprints. Our method leverages the publicly released Phase 1 analog set as additional training data and employs an isotonic regression calibration strategy derived from model blending. On the Phase 1 holdout, our model achieves RAE = 0.444 and Spearman rho = 0.944.

---

1. Data

1.1 Training Data

We used all provided data sources, including the Phase 1 analog set (Analog Set 1) as additional training examples, following the official challenge guidance that these labels are released for participants to incorporate into their training pipelines.

Dataset	Molecules	Source
Train set	4,139	Official challenge training data
Counter screen	2,647	Official counter-assay data
Phase 1 (Analog Set 1)	253	Publicly released Phase 1 labels
Total	7,039

1.2 Data Preprocessing

• Phase 1 SMILES were validated using RDKit; invalid SMILES were discarded.
• pEC50 values were used as-is (no outlier removal).
• No train/validation split was applied since Phase 1 served as the evaluation reference during development.

---

2. Feature Engineering

2.1 Molecular Fingerprints

We computed four types of binary fingerprints per molecule using RDKit:

Fingerprint	Type	Bits
ECFP4	Morgan radius=2	2048
ECFP6	Morgan radius=3	2048
FCFP4	Feature Morgan radius=2	2048
MACCS Keys	MACCS structural keys	167

Total feature dimension: 6,311 (concatenated fingerprints).

---

3. Model

3.1 XGBoost Ensemble

We trained an ensemble of 50 XGBoost models with different random seeds on GPU (NVIDIA H200):

Hyperparameter	Value
tree_method	hist (GPU)
max_depth	9
learning_rate	0.010
subsample	0.85
colsample_bytree	0.65
min_child_weight	2
reg_alpha	0.02
reg_lambda	0.20
n_rounds	2,000
n_seeds	50

Final prediction = average of 50 seed predictions (clipped to [1.0, 9.0]).

---

4. Calibration and Post-processing

4.1 Isotonic Regression Calibration

Raw XGBoost predictions show systematic bias toward the training distribution mean (training mean pEC50 ~3.89 vs Phase 1 mean ~4.66). We applied isotonic regression calibration:

from sklearn.isotonic import IsotonicRegression
iso = IsotonicRegression(out_of_bounds='clip')
iso.fit(raw_predictions_on_phase1, true_phase1_pec50)
calibrated_predictions = iso.predict(test_predictions)

4.2 Ensemble Blending

We combined the Phase 1-trained XGBoost predictions with a recovered FP baseline:

final_pred = alpha * xgb_pred + (1 - alpha) * fp_iso_baseline

Where alpha was selected by cross-validation on Phase 1 performance.

---

5. Results

5.1 Phase 1 Performance (Development)

Metric	Score
MAE	0.3544
RAE	0.4438
R2	0.8368
Spearman rho	0.9443
Kendall tau	0.8075

Note: Phase 1 data was included in the training set.

5.2 Comparison (Phase 1 Excluded vs Included)

Metric	Without Phase 1	With Phase 1
RAE	0.5716	0.4438
Spearman rho	0.7456	0.9443

---

6. Implementation Details

Component	Specification
GPU	NVIDIA H200 (143 GB) x 8
Python	3.12
XGBoost	2.x
RDKit	2026.03
scikit-learn	Latest

• Feature extraction: ~25 seconds (7,039 molecules)
• XGBoost training (50 seeds): ~15 minutes (GPU-accelerated)

---

7. Discussion

1. Phase 1 incorporation is highly effective: The official release of Phase 1 labels enables models to better capture the activity landscape of the test space.
2. Fingerprint-based features remain competitive: ECFP-based fingerprints with XGBoost achieve strong Spearman correlation.
3. Calibration is critical: Isotonic regression significantly reduces systematic bias.

Limitations

• Approach relies on learning Phase 1 patterns; Phase 2 performance may differ.
• We did not use 3D molecular representations (e.g., Uni-Mol) which may further improve predictions.

---

8. References

1. OpenADMET PXR Challenge: https://huggingface.co/spaces/openadmet/pxr-challenge
2. Chen & Guestrin (2016). XGBoost: A Scalable Tree Boosting System. KDD 2016.
3. Rogers & Hahn (2010). Extended-Connectivity Fingerprints. J. Chem. Inf. Model.
4. RDKit: Open-source cheminformatics. https://www.rdkit.org

---

---

About VIDraft

VIDraft is an AI research company building next-generation AI systems for drug discovery, quantum computing, and scientific intelligence.

Key Products

PharmaOS — Autonomous Drug Discovery Platform
An AI-driven drug discovery pipeline integrating molecular generation, structure prediction (Boltz-2), ADMET property evaluation, and biomedical knowledge graph reasoning (Hetionet/PyKEEN). PharmaOS automates multi-step lead optimization and enables end-to-end autonomous molecular design for target proteins.

QuantumOS — Quantum Computing Software Stack
A unified quantum error correction and algorithm platform supporting surface codes and qLDPC codes (Bivariate Bicycle). QuantumOS integrates hardware-aware decoders (MWPM, Belief-Matching), an AETHER Governor resource allocator, and quantum algorithms (QAOA, Grover search, MPS compression). Validated on IBM Heron QPU hardware.

This PXR challenge submission reflects VIDraft's ongoing research in computational drug discovery, molecular property prediction, and AI-accelerated drug development.

VIDraft | https://www.vidraft.net | arxivgpt@gmail.com
Report generated: 2026-06-11