OpenBioMed - An open-source agent platform jointly launched by Tsinghua AIR and SMTH Blog

What is OpenBioMed?

OpenBioMed is an open-source platform jointly launched by the Institute for Intelligent Industry (AIR) of Tsinghua University and SMTH Molecular, focusing on AI-driven biomedical research. It is a multimodal representation learning toolkit capable of processing various biomedical data, including molecules, proteins, and single cells. The platform provides over 20 tools and deep learning models, such as the BioMedGPT series, supporting a wide range of applications from traditional drug discovery tasks to multimodal challenges.

Main functions of OpenBioMed

• Multimodal data supportIt supports a variety of biomedical data, including molecular structures of small molecules, proteins, and single cells, transcriptomics, knowledge graphs, and biomedical texts.
• Unified data processing frameworkIt can easily load data from different biomedical entities and modalities and convert them into a unified format.
• Rich pre-trained modelsIt includes more than 20 deep learning models, such as BioMedGPT-10B, MolFM, CellLM, etc., which can be used for a variety of biomedical tasks.
• Diverse computing toolsMore than 20 computational tools have been developed, covering molecular property and structure prediction, molecular retrieval, molecular editing, and molecular design.
• Model prediction moduleIt discloses the parameters of the pre-trained model and provides use cases, making it easy to transfer to other data or tasks.
• Drug developmentIt can predict drug-target binding affinity, molecular properties, and drug response, accelerating new drug development.
• Multimodal understandingIt helps scientists find text descriptions related to molecules or proteins through cross-modal retrieval.
• Precision medicineCellLM-based cell type classification and single-cell drug sensitivity prediction drive personalized treatment.
• Intelligent Question AnsweringBioMedGPT can answer complex questions about molecules and proteins.
• Intelligent agent designWith a visual editing mode, researchers can easily access cutting-edge AI algorithms and tools by dragging and dropping to complete the design and development of intelligent agents.

OpenBioMed's technical principles

• Multimodal data processingOpenBioMed provides a flexible API for processing multimodal biomedical data, including small molecules, proteins, molecular structures of single cells, transcriptomics, knowledge graphs, and biomedical text.
• Deep learning modelsOpenBioMed integrates over 20 deep learning models, such as BioMedGPT-10B, MolFM, and CellLM. Through its advanced neural network architecture, it can handle tasks ranging from traditional AI drug discovery to emerging multimodal challenges.
• Pre-trained models and inferenceOpenBioMed provides ready-made pre-trained models and inference demos that have been trained on large-scale biomedical data and can be quickly transferred to users' own data or tasks.
• Tools and ApplicationsOpenBioMed has built more than 20 computational tools covering downstream tasks ranging from molecular property prediction to protein folding and cell type classification. These tools support a wide range of scenarios from basic research to clinical applications, such as generating molecular descriptions using the MolFM model or classifying cell types using the CellLM model.
• Intelligent Agents and WorkflowsOpenBioMed provides an easy-to-use interface for building workflows that connect multiple tools to develop agents based on Large Language Models (LLMs). These agents can simulate trial and error processes, helping researchers gain scientific insights in complex biomedical tasks.

OpenBioMed project address

• Project official website:openbiomed.pharmolix.com
• Github repository:https://github.com/PharMolix/OpenBioMed

Application scenarios of OpenBioMed

• Drug developmentOpenBioMed enables researchers to quickly screen for potentially effective drugs thanks to its powerful data processing capabilities and advanced machine learning algorithms.
• Multimodal understandingOpenBioMed supports cross-modal retrieval, helping scientists find textual descriptions related to molecules or proteins and enhancing their understanding of biomedical entities.
• Precision medicineIn the field of precision medicine, OpenBioMed uses the CellLM model to classify cell types and predict single-cell drug sensitivity, driving progress in personalized treatment.
• Knowledge Graph ConstructionOpenBioMed provides tools for building knowledge graphs, helping researchers to organically organize elements such as genes, proteins, drugs, and clinical symptoms to form a vast and sophisticated knowledge network.