How to Use Chai-1 Online

Chai-1: A Multi-Modal Foundation Model for Molecular Structure Prediction

Chai-1 (or chai1) is a new multi-modal foundation model for biomolecular structure prediction that has demonstrated state-of-the-art performance across a variety of tasks crucial for drug discovery. The model is designed to be highly versatile, excelling in areas such as protein-ligand and protein multimer structure prediction. Notably, Chai-1 is an openly accessible model, making it a viable and flexible alternative to other restricted models like AlphaFold3.

A key strength of Chai-1 is its ability to be prompted with experimental restraints, which can significantly boost prediction accuracy, especially for challenging binding complexes. These restraints can be derived from various wet-lab data sources like epitope mapping or cross-linking mass spectrometry. This integration of computational and experimental data makes Chai-1 a powerful tool for researchers.

Furthermore, Chai-1 can be run in a single-sequence mode without the need for multiple sequence alignments (MSAs), while still preserving most of its high performance. In this mode, Chai-1 has been shown to outperform other models like ESMFold. This capability is particularly advantageous when evolutionary information is sparse or unavailable. The model achieves a ligand RMSD success rate of 77% when given only the sequence and chemical composition, a result comparable to AlphaFold3.

How Chai-1 Works

Chai-1's architecture is a deep learning neural network with key improvements that allow it to handle multiple input modalities and complex prediction tasks. The model's design is based on the idea of incorporating diverse data to enhance its predictive power.

The model’s core functionality is centered on a series of input features and a multi-stage prediction process:

• Multi-modal Inputs: Chai-1 can take a wide range of inputs beyond just the protein sequence, including chemical information for ligands, and embeddings from a large protein language model.

• Single-Sequence Capability: To enable strong performance without MSAs, Chai-1 incorporates a separate input track with residue-level embeddings from a large protein language model. This feature allows the model to capture grammatical and structural information from the sequence itself.

• Experimental Constraints: The model has been trained with special features designed to mimic experimental constraints, such as inter-chain distance information. These features are trained with dropout to prevent the model from becoming overly reliant on them, but during inference, they can be used to guide predictions and achieve greater accuracy for difficult complexes.

• Prediction and Confidence: For protein-ligand complex prediction, Chai-1 uses a method that minimizes the RMSD between the predicted and native chain by referring to global docking scores, which can lead to better ligand RMSD success rates. The model’s predictions for protein-ligand interactions are highly accurate, and specifying the apo structure of a protein can boost its success rate to 81%.

What is Tamarind.bio?

Tamarind.bio is a pioneering no-code bioinformatics platform built to democratize access to powerful computational tools for life scientists and researchers. Recognizing that many cutting-edge machine learning models are often difficult to deploy and use, Tamarind Bio provides an intuitive, web-based environment that completely abstracts away the complexities of high-performance computing, software dependencies, and command-line interfaces.

The platform is designed to be accessible to biologists, chemists, and other researchers who may not have a background in programming or cloud infrastructure. Key features include a user-friendly graphical interface for setting up and launching experiments, a robust API for integration into existing research pipelines, and an automated system for managing and scaling computational resources. By handling the technical heavy lifting, Tamarind.bio empowers researchers to concentrate on their scientific questions and accelerate the pace of discovery.

Accelerating Discovery with Chai-1 on Tamarind Bio

The integration of Chai-1's advanced capabilities with Tamarind.bio's user-centric platform creates a powerful synergy that can significantly accelerate the drug discovery and research process.

• Unified Prediction and Flexibility: Chai-1's multi-modal nature allows it to perform a variety of tasks from a single model, including predicting protein-ligand and protein multimer interactions, which are critical for drug discovery. This is all available through a single, easy-to-use interface on Tamarind.bio.

• Rapid and Accurate Insights: By combining Chai-1's ability to run in fast single-sequence mode with Tamarind.bio's massive scalability, researchers can quickly screen thousands of protein-ligand interactions and get high-quality predictions without the need for extensive computational setup.

• Democratizing structural bioinformatics: The no-code platform makes state-of-the-art tools like Chai-1 accessible to a broader scientific community, including researchers in academia and small biotech companies who might not have access to dedicated computational resources. This democratization of access to advanced AI models for molecular structure prediction helps to accelerate fundamental research and drive innovation in therapeutic design.

How to Use Chai-1 on Tamarind Bio

Tamarind.bio makes using Chai-1 straightforward and efficient, regardless of your technical expertise. The no-code platform streamlines the entire workflow for molecular structure prediction.

Here is a simple, step-by-step guide for researchers to get started:

1. Access the Platform: Begin by logging in to the tamarind.bio website. The platform provides a central interface to access cutting-edge tools.

2. Select Chai-1: From the list of available computational models, choose the Chai-1 tool.

3. Specify Inputs: You can provide your input in various forms, including protein sequences (FASTA), chemical compositions for ligands (SMILES), and experimental constraint data. For predicting complexes, you can simply provide multiple sequences, and the platform will fold them together.

4. Configure Parameters: In a simple, graphical user interface, you can specify your prediction parameters. This includes choosing to use an MSA or single-sequence mode, and optionally adding experimental restraints. You can also set the number of recycles and diffusion steps, balancing accuracy with prediction time.

5. Submit and Monitor: With a single click, you can submit your job. The Tamarind platform handles the high-performance computing, parallelization, and GPU orchestration, saving you time and money. You can monitor the progress of your job directly from a user-friendly dashboard.

6. Analyze the Results: Once the job is complete, you will receive a comprehensive report with the predicted structures. You can explore interactive 3D visualizations of the structures directly in your browser. The platform also ensures that your inputs and outputs are securely stored in your private cloud, and that you retain ownership of all your data.