A naukri.com initiative
Hitconsultant
1d
199
Image Credit: Hitconsultant
NVIDIA Introduces La-Proteina: A Breakthrough in Atomistic Protein Generation via Partially Latent Flow Matching
- NVIDIA Research, in collaboration with the University of Oxford and Mila – Québec AI Institute, has introduced La-Proteina, a new method for atomistic protein design aiming to generate fully atomistic protein structures with their underlying amino acid sequences.
- La-Proteina utilizes a partially latent protein representation to model the backbone structure explicitly while capturing sequence and atomistic details through per-residue latent variables of fixed dimensionality, addressing challenges in explicit side-chain representations.
- The model is trained using a Variational Autoencoder (VAE) and a Partially Latent Flow Matching Model, achieving state-of-the-art performance in terms of designability, diversity, and structural validity, especially in scalability to large proteins.
- La-Proteina's architectural design involves neural networks based on efficient transformer architectures, with the denoiser network conditioning on interpolation times, totaling around 160M parameters, to enhance performance in atomistic protein generation.
Read Full Article
12 Likes
For uninterrupted reading, download the app