AlphaFold 2 검토

It's a deep learning network that achieved similar accuracy to AlphaFold 2 and has evolved into RoseTTAFold All-Atom, capable of modeling more complex biological assemblies including proteins, nucleic acids, small molecules, and metals.

RoseTTAFold demonstrated comparable accuracy to AlphaFold 2 in protein structure prediction, particularly in the CASP14 competition. Its 'All-Atom' version extends its capabilities beyond just proteins to a wider range of biomolecules, whereas AlphaFold 2 primarily focused on single protein chains and multimers, though AlphaFold 3 has expanded to predict interactions with DNA, RNA, and ligands.

It predicts protein structures from a single amino acid sequence using a protein language model, offering significantly faster prediction speeds compared to methods relying on multiple sequence alignments.

ESMFold offers comparable accuracy to AlphaFold 2 but with a major advantage in speed, as it doesn't require computationally expensive multiple sequence alignments (MSAs). AlphaFold 2 relies heavily on MSAs for its predictions. ESMFold is also available for free.

It is a fast, memory-efficient, and fully trainable open-source implementation of AlphaFold 2, designed to match its accuracy.

OpenFold aims to replicate and provide an open-source alternative to AlphaFold 2's capabilities and accuracy. While AlphaFold 2's code was open-sourced, OpenFold was built from the ground up to be a robust and generalizable implementation, addressing some of the challenges of using the original AlphaFold 2 code.

It's a comprehensive cloud service and framework for drug discovery that integrates various AI models, including protein language models and OpenFold, for biomolecular prediction and generation.

BioNeMo is a broader platform for drug discovery that includes protein structure prediction capabilities (e.g., via OpenFold and other large language models) rather than being solely a protein structure predictor like AlphaFold 2. It offers a framework for training and deploying large-scale biomolecular models, providing a more extensive toolkit for researchers.