Use Proto Online

Proto: A High-Level Programming Language for Generative Biology

Discover Proto, a high-level programming language for generative biology that encodes complex design campaigns across multiple modalities and scales—spanning DNA, RNA, proteins, ligands, and their interactive networks. By shifting the biological design process from manual, sequence-level trial-and-error to a functional and semantic level, Proto enables multi-objective, multi-modal biological engineering.

Core Primitives of the Proto Language

Proto simplifies biological design campaigns by breaking them down into four universal programming blocks:

• Sequences: Typed variables representing a discrete string of DNA, RNA, protein, or ligand elements.

• Constraints: Differentiable or non-differentiable scoring functions that map sequences to scalar desirability scores (e.g., GC content, structural predictions from AlphaFold, or epigenomic track predictions from Borzoi).

• Generators: Pretrained or pre-specified algorithmic models that propose biologically plausible sequence candidates by sampling from a learned distribution (e.g., ESM3, Evo 2, or ProteinMPNN).

• Optimizers: Search workflows that manage iterative design loops, compiling constraints into a unified energy function to steer the generator toward optimal, low-energy targets.

What is Tamarind Bio?

Tamarind Bio is a unified computational platform designed to streamline advanced machine learning and structural biology workflows directly from a web interface or programmatic API. Orchestrating state-of-the-art biological AI assets traditionally demands significant engineering overhead due to isolated software dependencies, hardware resource constraints, and unstandardized data formats.

The platform acts as a high-performance cloud gateway, natively provisioning model weights, managing environment isolation, and using cost-aware scheduling to scale workloads seamlessly across local and hosted GPU architectures. Tamarind Bio abstracts away the infrastructure complexity so researchers can focus entirely on biological creation.

How to Use Proto on Tamarind Bio

[User Specification] -> [Construct Scaffold] -> [Generators & Constraints] -> [Optimizer Compile] -> [Designed Sequences]

[User Specification] -> [Construct Scaffold] -> [Generators & Constraints] -> [Optimizer Compile] -> [Designed Sequences]

[User Specification] -> [Construct Scaffold] -> [Generators & Constraints] -> [Optimizer Compile] -> [Designed Sequences]

1. Select Your Interface: Choose how you want to interact with the language framework. Tamarind Bio supports a local/cloud Python API for programmatic wiring, a Graphical User Interface (GUI) featuring a drag-and-drop node canvas, and an Agentic Interface utilizing AI coding assistants to generate full scripts from natural language steering instructions.

2. Define Your Construct Scaffold: Map out your biological architecture by generating contiguous sequence regions called Segments (such as promoters, coding regions, or binders) inside a master structural template called a Construct.

3. Configure Your Generators and Constraints: Assign structural or genomic sequence generators to your variable segments. Layer on your desired combination of multi-objective filters, ranging from tissue-specific splice-site predictors (e.g., AlphaGenome) to biophysical interaction profiles.

4. Deploy a Search Optimizer: Pick an optimized approximate inference engine matching your construct's mathematical properties. Choose Metropolis-Hastings MCMC with simulated annealing for black-box energy landscapes, Rejection Sampling for embarrassingly parallel bioinformatic filtering, Gradient Descent for continuous differentiable logits, or Beam Search for autoregressive sequence prefix extensions.

5. Compile and Run: Compile your programmatic blocks into a unified energy-based distribution mapping desired function to sequence space. Submit the workflow to execute on Tamarind Bio's GPU architecture, which streams score trajectories, intermediate sequences, and 3D molecular structures straight back to your environment in real time.

Source