Genbio AI develops multiscale AI foundation models designed to decode and simulate human biology. The company is building toward an interactive, programmable virtual cell, referred to as an AI-Driven Digital Organism (AIDO). This technology aims to accelerate breakthroughs across medicine, biotech, and life sciences through advanced biological simulation.
Funding
Funding not disclosed
Founders
Product
Problem
Biological research is often fragmented, with specialized models addressing isolated tasks and data modalities. This siloed approach hinders the integration of knowledge across different biological scales, from molecular interactions to cellular behaviors and organismal phenotypes, making it difficult to uncover complex biological networks and accelerate discovery.
Solution
GenBio AI is developing the AI-Driven Digital Organism (AIDO), a unified system of multiscale foundation models designed to predict, simulate, and program biology across all levels. AIDO integrates diverse biological data, including DNA, RNA, proteins, and single-cell information, to create comprehensive representations of living systems. This approach aims to overcome the limitations of fragmented research by providing a holistic framework for understanding biological complexity. By leveraging advanced AI techniques, AIDO facilitates breakthroughs in drug discovery, bio-engineering, and disease prevention through predictive modeling and simulation.
Target Audience
The primary target audience includes researchers, biotechnologists, and healthcare professionals in fields such as drug discovery, genomics, proteomics, and personalized medicine.
Features
- **AIDO.DNA:** A 7 billion parameter DNA foundation model trained on 10.6 billion nucleotides across 796 species, supporting functional genomics and synthetic biology applications.
- **AIDO.RNA:** A 1.6 billion parameter RNA foundation model trained on 42 million non-coding RNA sequences, achieving state-of-the-art performance in RNA structure and function prediction.
- **AIDO.Protein:** A 16 billion parameter protein foundation model utilizing a Mixture-of-Experts (MoE) architecture for efficient training and inference, pretrained on 1.2 trillion amino acids.
- **AIDO.StructureTokenizer:** A VQ-VAE-based tokenizer for protein structures, designed for efficient 3D structure reconstruction and homology detection, enhancing protein language model performance.
- **AIDO.RAGPLM and AIDO.RAGFold:** Retrieval-augmented models for protein language modeling and structure prediction, improving accuracy and speed, particularly in low-MSA scenarios.
- **AIDO.Cell:** A series of single-cell foundation models (3M to 650M parameters) pretrained on 50 million human cells, processing entire transcriptomes for precise cellular context representation.
- **AIDO Platform:** An interactive toolkit enabling multiscale biological simulations, adaptation to sparse data via self-supervised learning, and pan-modal integration of biological data.
- **State-of-the-Art Performance:** Models achieve SOTA results in over 300 biological tasks, with large parameter counts (e.g., 70B+ for K2, 100B+ for Protein FM).