BioVAE

BioVAE provides a large-scale pre-trained latent variable language model for biomedical text mining, integrating pre-trained language models with deep generative modeling to model and generate biomedical literature.

Key Features:

• Domain-Specific Training: Trained on extensive volumes of biomedical literature using the OPTIMUS framework to capture biomedical text nuances.
• Latent Variable Model: Implements a latent variable language model that uses deep generative models to model complex data distributions in biomedical text.
• State-of-the-Art Performance: Demonstrates superior performance on multiple biomedical text mining benchmarks compared to other publicly available biomedical PLMs.
• Sentence Generation: Generates biomedical sentences with higher accuracy than the original outputs from the OPTIMUS framework.

Scientific Applications:

• Biomedical Literature Mining: Extracting relevant information and patterns from large collections of biomedical publications.
• Data Annotation and Curation: Assisting annotation and curation of biomedical datasets by producing contextually accurate text.
• Drug Discovery and Development: Supporting data analysis and text-generation tasks relevant to drug discovery and development research.

Methodology:

Trained via the OPTIMUS framework, integrating pre-trained language models (PLMs) with deep generative latent variable modeling on biomedical literature.