xTB2go

Stable Version: v6

Capabilities

• SMILES to 3D Structure Conversion
  Converts user-provided SMILES strings to 3D XYZ molecular structures using RDKit, enabling visualization and computational chemistry calculations.

• Geometry Optimization
  Performs geometry optimization of molecular structures using the xTB tight binding program, providing optimized geometries and energy data.

• Thermochemical Calculations
  Computes thermodynamic properties, including total energy, enthalpy, and free energy, using Hessian calculations. The results are extracted and presented clearly.

• Frequency Analysis
  Extracts vibrational frequencies from xTB's output files to identify the lowest frequencies, which are crucial for understanding molecular vibrations and confirming optimized geometries.

• Multiplicity and Spin State Control
  Automatically determines the appropriate multiplicity based on the number of electrons, allowing for accurate calculations of open-shell and closed-shell molecules.

• Solvent Effects Modeling
  Supports implicit solvation models (e.g., ALPB and GBSA) to study molecules in different solvent environments, enhancing the relevance of calculations to experimental conditions.

• Customizable Calculation Parameters
  Provides users with options to customize charge, method, and solvent model settings, allowing for tailored quantum chemical calculations.

• Dynamic Visualization of Molecules
  Integrates with py3Dmol for interactive 3D visualization of molecular structures directly in the web app interface.

• Downloadable Calculation Results
  Offers easy download of calculation results, including energies, optimized geometries, and frequencies, in a user-friendly .txt format.

• User-Friendly Streamlit Interface
  Utilizes a simple, interactive web interface built with Streamlit, making advanced quantum chemical calculations accessible to users without a steep learning curve.

• Error Handling and User Guidance
  Provides real-time feedback and error messages to guide users in correcting input errors or adjusting calculation parameters for successful runs.

• Advanced Computational Chemistry Tools Integration
  Seamlessly integrates with xTB for advanced semi-empirical methods, supporting a range of calculations from basic energy evaluation to complex thermochemical properties and conformational analyses.

Dependencies

• streamlit
• subprocess
• os
• rdkit
• py3Dmol
• openai

In the works:

xTB calculator powered by large language model. Current version uses OpenAI API.

• Version 1: One LLM assistant --> gets xTB command line from prompt & SMILES from PubChem
• Version 2: Two LLM assistants --> Second one parses the xTB output file to get energies and properties
• Current working version: pubchem_xtb_LLM_chunks_separated_prompt.py

Ongoing work

1. Debugging experimental examples, some are still causing errors.

2. Experiment with system prompt to improve xtb results analysis.

3. Clean up output.

4. Add functionality to choose different softwares (e.g. Gaussian) and generate input files.

To-Do List

1. Conformer Search with CREST
   Implement conformer search functionality using CREST for more accurate conformations.

2. Charge / Multiplicity Combinations
   Add support for specifying multiplicity in addition to charge for various molecular species.

3. Hessian Calculation
   Perform Hessian calculations to obtain and print out thermochemistry data. Status: Done

4. Ionization Potentials and Electron Affinities
   Introduce an option to calculate ionization potentials and electron affinities for molecules.

5. Additional Outputs Done (This can be done by parsing the output with LLM) Add options to print:

   • HOMO-LUMO Gap
   • Thermochemistry Data (if available)
   • Output format in JSON / csv

6. Run multiple jobs (from .csv file) and allow downloading csv output file

7. Connect to PubChem API – Similarity search Done

8. Connect to ChemSpider: waiting for website to update APIs

9. Expand on the LLM Prompt: Notes on what types of agents can we build.

10. ML integration (think more about this)

11. Generative AI (check notes from Google Drive)

12. Improve on Socratic LLM agent. Implement the table from the manuscript - monitor improvement in the LLM. Here we have some working examples with different tasks. To do: Improve on system prompts, connect agents with each other.

13. Experimental: Can we link Socratic LLM to xtb calculator? Aim: propose and run the molecules then report back with analysis and propose further prompts.