Tools, Workflows and Code

AcidAmine Descriptor Predict

This repository contains the necessary code for work done in the paper 'Rapid Prediction of Conformationally-Dependent DFT-Level Descriptors using Graph Neural Networks for Carboxylic Acids and Alkyl Amines' 

AcidAmine Descriptor Predict Github

Publication:

Haas, B.; Hardy, M. A.; Sowndarya, S. S.; Adams, K.; Coley, C. W.; Paton, R. S.; Sigman, M. S. Rapid Prediction of Conformationally-Dependent DFT-Level Descriptors using Graph Neural Networks for Carboxylic Acids and Alkyl Amines. Digit. Discovery 2025, 4, 222–233. https://doi.org/10.1039/D4DD00284A

---

AgentDrug

Molecular editing—modifying a given molecule to improve desired properties—is a fundamental task in drug discovery. While LLMs hold the potential to solve this task using natural language to drive the editing, straightforward prompting achieves limited accuracy. AgentDrug is anan agentic workflow that leverages LLMs in a structured refinement process to achieve significantly higher accuracy. AgentDrug defines a nested refinement loop: the inner loop uses feedback from cheminformatics toolkits to validate molecular structures, while the outer loop guides the LLM with generic feedback and a gradient-based objective to steer the molecule toward property improvement. 

AgentDrug GitHub

Publication:

Le, K.; Hua, T.; Chawla, N. V. AgentDrug: Utilizing Large Language Models in an Agentic Workflow for Zero-Shot Molecular Editing Conf. Emp. Meth. Natural Lang. Proc (EMNLP2025) 2025 24448–24458. https://aclanthology.org/anthology-files/anthology-files/pdf/findings/2025.findings-emnlp.1328.pdf

---

AIMNet2

This package integrates the powerful AIMNet2 neural network potential into your simulation workflows. AIMNet2 provides fast and reliable energy, force, and property calculations for molecules containing a diverse range of elements.

AIMNet2

Publication:

Anstine, D.M., Zubatyuk, R. and Isayev, O., 2025. AIMNet2: a neural network potential to meet your neutral, charged, organic, and elemental-organic needs. Chem. Sci. 2025, 16, 10228-10244. https://doi.org/10.1039/D4SC08572H

Tutorial Material:

AIMNet2 Tutorial

UMA Tutorial

---

AQME

AQME is an ensemble of automated QM workflows, including: 1) RDKit- and CREST-based conformer generator and ready-to-submit QM input files starting from individual files or databases, 2) post-processing of QM output files to fix extra imaginary frequencies, unfinished jobs and error terminate

AQME

Publication:

Alegre-Requena, J. V.; Sowndarya, S.; Pérez-Soto, R.; Alturaifi, T.; Paton, R. AQME: Automated Quantum Mechanical Environments for Researchers and Educators. Wiley Interdiscip. Rev. Comput. Mol. Sci. 2023, 13, e1663 doi.org/10.1002/wcms.1663

Tutorial Material:

AQME Tutorial

---

Auto-QChem

Auto-QChem is an automatic, high-throughput and end-to-end DFT calculation workflow that computes chemical descriptors for organic molecules. Tailored toward users without extensive programming experience, Auto-QChem has facilitated more than 38 000 DFT calculations for 17 000 molecules as of January 2022. Starting from string representations of molecules, Auto-QChem automatically (a) generates conformational ensembles, (b) submits and manages DFT calculations on a high-performance computing (HPC) cluster, (c) extracts production-ready features that are suitable for statistical analysis and machine learning model development, and (d) stores resulting calculations in a cloud-hosted and web-accessible database

Auto-QChem Github

Publication:

Żurański, A.M.; Wang, J.Y.; Shields, B.J.; Doyle, A.G. Auto-QChem: an automated workflow for the generation and storage of DFT calculations for organic molecules. React. Chem. Engin. 2022, 7, 1276-1284 doi.org/10.1039/D2RE00030J

---

Bandit-Optimization

This code uses reinforcement learning, specifically the multi-armed bandit approach, for reaction optimization. It demonstrates data-efficient learning at high accuracies and has unique functionalities.

Bandit Optimizer Code

Publication:

Wang JY, Stevens JM, Kariofillis SK, Tom MJ, Golden DL, Li J, Tabora JE, Parasram M, Shields BJ, Primer DN, Hao B. Identifying general reaction conditions by bandit optimization. Nature. 2024, 626, 1025-1033. doi.org/10.1038/s41586-024-07021-y

---

Cascade

CASCADE stands for ChemicAl Shift CAlculation with DEep learning. It is a stereochemistry-aware online calculator for NMR chemical shifts using a graph network approach developed at Colorado State University. Molecular input can be specified as SMILES or through the graphical interface. An automated workflow executes 3D structure embedding and MMFF conformer searching. The full ensemble of optimized conformations are passed to a trained graph neural network to predict the NMR chemical shift (in ppm) for each carbon.

Cascade Download

Cascade Web Interface

Publication:

Guan, Y.; Sowndarya, S. V. S.; Gallegos, L. C.; St. John P. C.; Paton, R. S. Real-time prediction of ¹H and ¹³C chemical shifts with DFT accuracy using a 3D graph neural network Chem. Sci. 2021, 12, 12012-12026 doi.org/10.1039/D1SC03343C

---

ChemOrch

ChemOrch is a framework that synthesizes chemically grounded instruction-response pairs through a two-stage process: task-controlled instruction generation and tool-aware response construction. ChemOrch enables controllable diversity and levels of difficulty for the generated tasks, and ensures response precision through tool planning and distillation, and tool-based self-repair mechanisms. 

ChemOrch GitHub

Publication:

Huang, Y.; Jiang, Z.; Luo, X.; Guo, K.; Zhuang, H.; Zhou, Y.; Yuan, Z.; Sun, X.; Schleinitz, J.; Wang, Y.; Zhang, S.; Surve, M.; Chawla, N. V.; Wiest, O.; Zhang, X. ChemOrch: Empowering LLMs with Chemical Intelligence via Groundbreaking Synthetic Instructions NeurIPS 2025, 39, 6578

---

DA DataExtraction

A series of Jupyter notebooks and instructions for extracting a dataset of Diels–Alder reactions.

DA DataExtraction GitHub

---

DBStep

DBStep is a python package for obtaining DFT-Based Steric Parameters from 3-dimensional chemical structures. It can parse the outputs from most computational chemistry programs and other common molecular structure file formats. Steric properties can either be obtained exactly or by using a Cartesian grid, the latter approach being amenable to the featurization of a molecular isodensity surface (DBSTEP can process wavefunction files) rather than using classical atomic radii. Currently, traditional Sterimol parameters (L, Bmin, Bmax) and percent buried volume parameters are implemented, as well as our novel steric parameter vectors Sterimol2vec and vol2vec. This package is designed for use on the command line or alternatively implemented in a Python script for use in a computational workflow to collect steric parameters

DB Step Download

Publication:

Luchini, G.; Patterson, T.; Paton, R. S. DBSTEP: DFT Based Steric Parameters. 2022, DOI: 10.5281/zenodo.4702097

Tutorial Material:

DB Step Tutorial

Overview of modern steric parameters

---

DESP: Double-Ended Synthesis Planning with Goal-Constrained Bidirectional Search

This repository contains code for DESP (Double-Ended Synthesis Planning), which applies goal-constrained bidirectional search to computer-aided synthesis planning. DESP is designed to propose a synthesis plan towards a given target molecule under the user-specified constraint of using one or more specific starting materials. Read more about the algorithm in our preprint.

DESP Github

Publication:

Yu, K.; Roh, J.; Li, Z.; Gao, W.; Wang, R.; Coley, C. W. Double-ended synthesis planning with goal-constrained bidirectional search. Adv. Neural Inf. Process. Syst. (NeurIPS) 2025, 112919–112949. https://nips.cc/virtual/2024/poster/95604

---

Desulfonylative Fluorination

Modeling scripts and data associated with the Sigman-Sanford lab collaboration on desulfonylative fluorination of heteroaromatics. DOI: pending

Desulfonylative Fluorination GitHub

Publication:

Hall, J. R.; Romer, N. P.; Spiller, T.; Sigman, M. S.; Sanford, M. S., 2025. Pd-Catalyzed Desulfonylative Fluorination of Electron Deficient (Hetero) Aryl Sulfonyl Fluorides. Chem. Sci. 2025, 16, 18936–18941. https://doi.org/10.1039/D5SC00912J

---

EDBO+

EDBO+ is a multi-objective reaction Bayesian optimization platform that builds on the previously published Bayesian optimizers EDBO. The web-based application incorporates features such as condition modification on the fly and data visualization.

EDBO+ Web Interface

EDBO+ Github

Publications:

EDBO+ Torres, J. A. G.; Lau, S. H.; Anchuri, P.; Stevens, J. M; Tabora, J. E; Li, J.; Borovika, A.; Adams, R. P; Doyle, A.G. A Multi-Objective Active Learning Platform and Web App for Reaction Optimization. J. Am. Chem. Soc. 2022, 144, 1999-2007. doi/10.1021/jacs.2c08592

EDBO Shields, B.J.; Stevens, J.; Li, J.; Prarasram, M.; Damani, F.; Martinez Alvaro, J., Janey, J. Adams, R.P., Doyle, A. Bayesian Reaction Optimization as A Tool for Chemical Synthesis. Nature 2021, 590, 89-96. doi.org/10.1038/s41586-021-03213-y

Tutorial material:

EDBO+ Tutorial

Intro to Bayesian Optimization

"Over the Arrow’ Optimization

---

GoodVibes

GoodVibes is a Python program to compute thermochemical data from one or a series of electronic structure calculations. It has been used since 2015 by several groups, primarily to correct the poor description of low frequency vibrations by the rigid-rotor harmonic oscillator treatment. The current version includes thermochemistry at variable temperature/concentration, various quasi-harmonic entropy and enthalpy schemes, automated detection of frequency scaling factors, D3-dispersion corrections calculations, Boltzmann averaging, duplicate conformer detection, automated tabulation and plotting of energy profiles, and error checking.

GoodVibes

---

HT TSs Opt

This repository contains scripts for high-throughput generation, optimization, and featurization of TSs and catalytic cycle intermediates, along with scripts for MLR and active learning modeling and Excel spreadsheets with input data.

HT TSs Opt Github

Publication:

Gallarati, S.; Bucci, E. M.; Doyle, A. G.; Sigman, M. S. Transferable enantioselectivity models from sparse data. Nature 2026, 651, 637–646. https://www.nature.com/articles/s41586-026-10239-7

---

LabSafety Bench

Artificial Intelligence (AI) is revolutionizing scientific research, but its growing integration into laboratory environments brings critical safety challenges. As large language models (LLMs) and vision language models (VLMs) are increasingly used for procedural guidance and even autonomous experiment orchestration, there is a risk of an "illusion of understanding" where users may overestimate the reliability of these systems in safety-critical situations.

LabSafety Bench is a comprehensive evaluation framework designed to rigorously assess the trustworthiness of these models in laboratory settings. The benchmark includes two main evaluation components:

• Multiple-Choice Questions (MCQs):
  A set of 765 questions derived from authoritative lab safety protocols, comprising 632 text-only questions and 133 multimodal questions.

• Real-World Scenario Evaluations:
  A collection of 404 realistic laboratory scenarios that yield a total of 3128 open-ended questions, organized into:

  • Hazards Identification Test: Models identify all potential hazards in a given scenario.
  • Consequence Identification Test: Models predict the outcomes of executing specific hazardous actions.

LabSafety Bench GitHub

Publication:

Zhou, Y.; Yang, J.; Huang, Y.; Guo, K.; Emory, Z.; Ghosh, B.; Bedar, A.; Shekar, S.; Liang, Z.; Chen, P. Y.; Gao, T.; Geyer, W.; Moniz, N.; Chawla, N. V.; Zhang, X. Benchmarking LLMs on safety issues in scientific labs. Nat. Mach. Intell. 2026, 8, 20–31. https://doi.org/10.1038/s42256-025-01152-1

---

LLM Extraction Chem

In the realm of chemistry, literature texts elucidating chemical reactions are crucial for tasks such as yield prediction, reaction prediction, and reaction condition recommendation. However, extracting structured data from these texts is challenging. Our paper explores the use of Large Language Models (LLMs) to process unstructured chemical reaction data. We manually curated a dataset to fine-tune LLMs, including LLAMA-2 and GPT models, and employed techniques like prompt-tuning and fine-tuning. Our results show that prompt-tuning GPT-4 yields the highest accuracy. We also introduced a verification step using external APIs to authenticate extracted data. This work demonstrates the potential of LLMs in enhancing data-driven chemistry.

LLM Extraction Chem GitHub

Publication:

Huang, X.; Surve, M.; Liu, Y.; Luo, T.; Wiest, O.; Zhang, X.; Chawla, N. V. Application of large language models in chemistry reaction, data extraction, and cleaning. CIKM’24: Proc.33rd ACM Intl.Conf. Inf. Knowl. Manag. 2024, 33, 3797–3801. https://doi.org/10.1145/3627673.3679874

---

Maxbridge

Maxbridge is a web-based deterministic graphing program that permits the identification of the maximally bridged ring (or rings) for any molecule using the Chemistry Development Kit (CDK) software library

Maxbridge

Publication:

Marth, C.J., Gallego, G.M., Lee, J.C., Lebold, T.P., Kulyk, S., Kou, K.G.M., Qin, J., Lilien, R. and Sarpong, R., 2015. Network-analysis-guided synthesis of weisaconitine D and liljestrandinine. Nature, 528(7583), pp.493-498. doi.org/10.1038/nature16440

---

Molcomplex

This package is developed in a collaboration of the Paton and Sarpong groups. It Implements a variety of complementary metrics for molecular complexity and synthetic accessibility.

Molcomplex code

Molcomplex web interface

---

Multi-Linear Regression

Multilinear regressions (MLR) is a widely used tool in predictive chemistry and the elucidation of mechanism. C-CAS researchers developed a variety of tools for conducting and visualizing MLR analyses, the rapid generation of chemically relevant features, and feature databases such as Kraken (see above).

The different components of the MLR workflows are:

1) Workflow scripts for automated collection of molecular properties, as well as atom- and bond-level properties for a conserved moiety of interest from Gaussian jobs. Post-processing allows for the collection of condensed descriptors for conformational ensembles. https://doi.org/10.5281/zenodo.19560208

2) Open-source Python package for generation of SMART probe ensembles and calculation of SMART molecular descriptors https://doi.org/10.5281/zenodo.19560206

3) Workflow for linear modeling, primarily driven by bidirectional stepwise MLR, as well as threshold analysis and tools for feature curation https://doi.org/10.5281/zenodo.19560132

4) Workflow for classification of chemical compounds as active or inactive based on experimental outputs and a set of previously computed descriptors based on sci-kit learn's DecisionTreeClassifier. It yields models that can be both interpretable and predictive. https://doi.org/10.5281/zenodo.19560178

The analysis section also includes scripts for using HoloViews (Bokeh backend) to generate interactive plots https://doi.org/10.5281/zenodo.19560202

---

Multi-Threshold Analysis

This repository contains a workflow for classification of chemical compounds as active or inactive based on experimental outputs and a set of previously computed descriptors. Classification is performed via sci-kit learn's DecisionTreeClassifier and yields models that can be both interpretable and predictive.

---

One Step Retro Failure Mode

This repository contains code and scripts for analyzing and quantifying failure modes of one-step retrosynthesis models, as described in the publication below. Please refer to the paper for detailed methodology and results.

One Step Retro Failure Mode GitHub

Publication:

Tran, S. B.; Roh, J.; Coley, C. W. Quantifying the Failure Modes of Current One-Step Retrosynthesis Models. Chem. Sci. 2026, 17, https://doi.org/10.1039/d6sc01323f

---

PericyclicTL

Data generation notebooks for pericyclic datasets. These notebooks are for regenerating reaction data used in an upcoming publication.

Data generation notebooks for Cope and Claisen, Ene, and Nazarov datasets are here. For Diels–Alder reaction, see https://github.com/angusketo/DA_DataExtraction

PericyclicTL GitHub

Proto-Yield

Proto-Yield is an encoder-agnostic prototype network that models reactions as occurring in one of three yield regimes: high, medium, or low. Without access to full reaction processes, Proto-Yield learns to infer latent regimes and their associated yield distributions from noisy, incomplete training data.

Proto-Yield GitHub

Publication:

Guo, K.; Liu, Z.; Guo, Z.; Nan, B.; Isayev, O.; Chawla, N. V.; Wiest, O.; Zhang, X. Proto-Yield: An Uncertainty-Aware Prototype Network for Yield Prediction in Real-world Chemical Reactions 34th ACM Intl. Conf. Inf. Knowl. Man. (CIKM’25) 2025, 791–805. https://doi.org/10.1145/3746252.3761323

---

ReactionTeam

ReactionTeam, is composed of specialized expert models, each trained to capture a distinct type of electron redistribution pattern in reaction, and a ranking expert that evaluates and orders the generated predictions

ReactionTeam GitHub

Publication:

Guo, T.; Ma, G.; Guo, K.; Chen, X.; Nan, B.; Pi, S.; Chawla, N.; Wiest, O.; Zhang, X. ReactionTeam: Teaming Experts for Divergent Thinking Beyond Typical Reaction Patterns IEEE BigData2025 2025 10.1109/BigData66926.2025.11402365.

---

REyes (Reciprocal Eyes)

This repository contains a series of scripts designed to process diffraction data, generate heatmaps, identify key targets, and manage navigation files for SerialEM. The scripts are structured to be executed in the order outlined below, ensuring that new blocks of data are processed systematically while maintaining data integrity.

REyes GitHub

Publication:

Eremin, D. B.; Jha, K. K.; Delgadillo, D. A.; Zhang, H.; Foxman, S. H.; Johnson, S. N.; Vlahakis, N. W.; Cascio, D.; Lavallo, V.; Rodriguez, J. A.; Nelson, H. M. Spatially Aware Defraction Mapping Enables Fully Autonomous MicroED. J. Am. Chem. Soc. 2025, 147, 4299–42310. https://doi.org/10.1021/jacs.5c10751

---

RiskLab

When multiple LLM agents interact — negotiating prices, relaying information, or making collective decisions — new risks emerge from the interaction itself, not from any single agent's failure. Agents may silently collude on prices, conform to majority opinion, drift meaning across handoffs, or rigidly follow outdated instructions. These phenomena mirror well-studied human social dynamics (groupthink, cartel behavior, telephone-game distortion), yet no existing toolkit treats them as first-class, measurable objects.

RiskLab fills this gap. It provides a controlled experimental framework where every risk scenario is fully specified by a topology – environment – protocol – agent – task quintuple, making emergent risks programmable, reproducible, and quantitatively evaluable.

RiskLab GitHub

Publication:

Yu, J.; Wang, W.; Huang, Y.; Wang, Y.; Zhou, Z.; Chen, X.; Liu, Y.; Wang, W.; Zhang, X. RISKLAB: A Controlled Toolkit for Probing Emergent Risks in LLM-Based Multi-Agent Systems. Proc. 64th Ann. Meet. ACL, 2026 167–177. https://doi.org/10.18653/v1/2026.acl-demo.17

---

Rxnpredict

Predicting reaction performance using machine learning

Rxnpredict-2 GitHub

---

SPiCE: Symmetry-Preserving Conformer Ensemble Networks for Molecular Representation Learning

SPiCE learns molecular properties from conformer ensembles while preserving joint equivariance to geometric transformations of individual conformers and permutations of the ensemble. Key components:

• Geometric Mixture-of-Experts (GMoE): Separate per-atom routers for scalar and vector features using Gumbel softmax
• Hierarchical Ensemble Encoding: Cross-attention between topology (GIN) and conformer representations
• Upcycling: Progressive expert specialization from a shared initialization

SPiCE GitHub

Publication:

Zhu, Y.; Shi, Y.; Chen, Y.; Sun, F. Sun, Y.; Wang, W. Symmetry-Preserving Conformer Ensemble Networks for Molecular Representation Learning NeurIPS 2025, 39.

---

Threshold

Python tool to assess data for single-parameter thresholds

Threshold

---

wSterimol

wSterimol is an automated computational workflow which computes multidimensional Sterimol parameters. For flexible molecules or substituents, the program will generate & optimize a conformational ensemble, and produce Boltzmann-weighted Sterimol parameters. It has been developed as a PyMol plugin.

wSterimol GitHub

---

yield-rxn

Code for the paper: Graph Neural Networks for Predicting Chemical Reaction Performance

yield-rxn GitHub