BloClaw
Multi-modal agent workspace for controlling bioinformatics and chemoinformatics computational pipelines.
BloClaw, a multi-modal agent workspace dedicated to scientific research, is designed to address the various control challenges that arise when large language models handle bioinformatics and cheminformatics computational pipelines. Much like a skilled chemistry lab assistant meticulously mixing complex reagents and arranging tools according to the researcher's instructions, this system reliably mediates LLM tool calls and ensures sophisticated structured responses through the XML-Regex Dual-Track Routing Protocol. Existing agent frameworks suffered from a critical limitation: they frequently failed in environments requiring precise control, such as scientific computational pipelines, due to the use of error-prone JSON-based tool calls. Furthermore, they were unable to reliably extract data visualization generated during the analysis process outside the agent environment, which increased the researcher's cognitive load. BloClaw addresses this issue by capturing and mapping visualization results, such as those from Plotly or Matplotlib, in real-time through Python Runtime State Interception. This allows researchers to intuitively observe experimental data without opening a browser or external explorer. Biotechnology researchers can use BloClaw to safely perform tasks locally, from generating 2D molecular images based on RDKit to predicting 3D protein structures in conjunction with ESMFold. For example, by inputting the amino acid sequence of a specific target protein and running the ESMFold prediction pipeline, a 3D structure PDB file can be obtained locally, and the physicochemical properties of candidate compounds can be calculated using RDKit and immediately visualized. By providing a Zero-Trust privacy environment where sensitive material data is not leaked to external clouds, it can be immediately adopted as a secure AI research assistant in new drug development laboratories or the pharmaceutical industry.
💻 System Requirements
Minimum 8GB NVIDIA GPU VRAM (12GB+ recommended for local ESMFold folding)
Minimum 10GB or more (including molecular informatics weights such as RDKit and ESMFold)
⚡ Installation
### 4-1. Quick Start
```bash
git clone https://github.com/qinheming/BIoClaw.git
cd BIoClaw && poetry install
```
### 4-2. Detailed Installation
```bash
# Enter the Poetry virtual environment and run the main module
poetry run python main.py
```🧬 Bio Use Cases
🔬 Render 2D structures and calculate properties for a large number of candidate molecules.
Using BloClaw to call the RDKit package, parse 1,000 molecules in SMILES format (parameter: 2D structure image generation options), and quantify physicochemical properties such as molecular weight (MW) and partition coefficient (LogP) for each compound in the local runtime within 10 seconds. This data is then used in subsequent virtual screening to improve the speed of preclinical design.
🧬 Integrate ESMFold for initial prediction and visualization of protein 3D structures.
By linking a protein sequence analysis tool with ESMFold, predict the 3D folded structure of a novel drug target protein (based on a sequence length of 400 amino acids, taking approximately 15 seconds on a single GPU). The generated PDB coordinates are then dynamically rendered within Matplotlib 3D plots and a local viewport to maximize the efficiency of identifying the target-ligand binding site (Active Site).
📊 Analyze biological activity data and create dynamic graph mappings.
Search for compound activity indicators (IC50) in the PubChem database and organize the dataset from the latest literature using a RAG model. Then, using the Python runtime interception feature, generate a Plotly dynamic box plot graph in real-time to visually classify the distribution of the top 5% of candidate molecules with excellent binding activity.
📝 Update Notes
No update notes yet.
🧪 Related Code of Life
No related Code of Life posts yet.