Skip to content
View MauricioCafiero's full-sized avatar
💭
Building protein tools and AI Agents
💭
Building protein tools and AI Agents

Block or report MauricioCafiero

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
MauricioCafiero/README.md

Latest additions!

InflectionLM - Output and Token Visualization

A tool that visualizes "inflection points" in LLM generation by generating multiple independent output paths and highlighting low-confidence tokens — moments where the model's output could easily have diverged.

  • Generates 3 independent responses per prompt using GEMMA 4 with Top-P/Top-K sampling for diversity
  • Highlights tokens with probability < 0.6 in red, marking them as inflection points
  • Interactive Gradio GUI with toggleable per-token probability scores and light/dark mode
  • See the repo!
  • Try the HuggingFace demo!

dock_assist - AI-assisted blind docking

A conversational CLI agent (powered by Ollama) that takes a plain-English request like "dock dopamine into SULT1A3" and autonomously runs the full workflow: PDB lookup, blind binding-site detection, and AutoDock Vina docking — no prior knowledge of the binding site required.

  • Pure-Python buriedness grid scan for pocket detection — no external pocket tool needed
  • Proximity fallback: auto-corrects off-target poses using co-crystallized ligand proximity
  • Validated on DUD-E receptors (HMGCR, ADRB1, ADRB2, MAOB, DRD2) — true binding site is #1 pocket on all five
  • See the repo!

FAO_MOLPROP_CLI - Fragment-based Optimization on the command line

Adversarial molecule optimization from the command line: an Ollama main model designs molecules to bind a protein target while an OpenAI/Anthropic adversary critiques each proposal, with all steps written to a timestamped Markdown report.

MoDrAg_CLI - Command-Line MoDrAg

A full-featured CLI version of the MOdular DRug design AGent with rich terminal output, IC50 prediction, and an easy node-integration system for adding new tools.

  • Molecular tools: SMILES conversion, similarity search, 3D structure generation, pharmacophore analysis
  • Protein tools: UniProt/PDB lookup, ChEMBL bioactives, AutoDock Vina docking, LightGBM IC50 prediction
  • Setup script to create a modrag shell alias for use from anywhere
  • See the repo!

Quantum computing intro

Find background on the quantum mechanics of quantum computing as well as introductions to quantum computing with Q#, Cirq and Qiskit!

Use the MLIP-based docking code, UMADock

  • Dock molecules is a selction of 6 proteins using the UMA MLIP for all energy evaluations.
  • Includes ligand desolvation and strain energy.
  • See the repo here!

Pinned Loading

  1. dock_assist dock_assist Public

    AI-assisted blind docking for SMILES strings on any protein

    Python 1 1

  2. MoDrAg_CLI MoDrAg_CLI Public

    Command Line Modrag

    Python

  3. FAO_MOLPROP_CLI FAO_MOLPROP_CLI Public

    CLI version of Fragment-based AI-assisted Optimisation of Molecular Properties

    Jupyter Notebook

  4. UMADock UMADock Public

    Docking molecules in protein binding sites using Meta's UMA MLIP as the energy scoring function. Also runs with an AI Agent.

    Jupyter Notebook

  5. CafChemQuantum CafChemQuantum Public

    Quantum Computing practice / teaching code in Q#, Qiskit and Cirq

    Jupyter Notebook

  6. Gaussian16-Scripts-and-Functions Gaussian16-Scripts-and-Functions Public

    Useful bash functions and scripts for running Gaussian 16 on Linux

    Shell 4