We try, so the community can optimize.

---

AGENTi V3: Machine Learning Framework

Overview

AGENTi V3 is a deployment package designed for deep-learning-based chemical analysis. It utilizes an ensemble of YDF Decision Forests and TensorFlow Autoencoder architectures to provide both regression scores (A-Score) and anomaly detection for chemical compounds provided in SMILES format.

1. Technical Requirements

To ensure consistent results, the following environment is required:

• Python: 3.9 - 3.12
• Primary Libraries: tensorflow, tensorflow-decision-forests, rdkit, scikit-learn, pandas, joblib.
• Hardware: Compatible with CPU-only environments (standard for many chemical informatics workflows).

2. Installation & Setup

Step 1: Download & Extract Download agenti_v3.zip and extract its contents into your working directory. Recommend running the following command in your terminal to automatically download, extract, and clean up the files in your current working directory:

wget -qO- https://githubusercontent.com > agenti_v3.zip && unzip -q agenti_v3.zip && rm agenti_v3.zip

Step 2: Initialize Environment You can use the automated setup script which handles the registration of kernels and all dependencies:

# Give execution permissions and run the installer
chmod +x setup.sh
./setup.sh

Alternatively, manually install via pip: pip install -r requirements.txt

3. Execution Guide

A. Single Molecule Inference

Use agenti_run.py to analyze a single compound. This script returns the A-Score and a flag indicating if the molecule is considered an anomaly.

python3 agenti_run.py --smiles "CC(=O)OC1=CC=CC=C1C(=O)O"

B. High-Throughput Batch Processing

For processing multiple molecules simultaneously, use agenti_v3_batch_production.py.

From Text Files: Input should be a .txt file with one SMILES string per line.

python3 agenti_v3_batch_production.py --input my_compounds.txt --output results.csv

From CSV Files: Specify the column name containing the SMILES strings.

python3 agenti_v3_batch_production.py --input library.csv --column SMILES_STR --output results.csv

4. Package Components

File/Folder	Description
agenti_run.py	The main entry point for single inference tasks.
agenti_v3_batch_production.py	Optimized script for large datasets (TXT/CSV).
agentI_reg_v3_minmax/	YDF Decision Forest weights for regression scoring.
agentI_anom_v3_minmax/	YDF Decision Forest weights for anomaly classification.
autoencoder_v3_minmax.keras	Trained TF Autoencoder model for feature reconstruction.
scaler_v3_minmax.pkl	Serialized normalization parameters for input features.

5. Troubleshooting

If you encounter a RuntimeError regarding unregistered kernels, ensure that tensorflow_decision_forests is installed and imported at the top of your execution environment. The provided scripts are pre-configured to handle this automatically.

6. Citation

Mithony Keng, Kenneth M Merz. AGENTi: APCI-ESI Analyte Ionization Fidelity Pre-Screening Workflow using Multimodel Inference. ChemRxiv. 02 July 2026. DOI: https://doi.org/10.26434/chemrxiv.15005565/v1