Learned protein embeddings for machine learning

By Kevin Kaichuang Yang, Zachary Wu, Claire N. Bedbrook, Frances H. Arnold, Jonathan D. Wren. 2018

132 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Molecular Biology, Representation & Self-Supervised Learning, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Sergey Levine
2. Pieter Abbeel
3. Jian Peng
4. Chelsea Finn
5. Yan Duan
6. Xi Chen
7. Yuxi Li
8. Igor Mordatch

In Molecular Biology

1. Kevin Kaichuang Yang
2. Zachary Wu
3. Frances H. Arnold
4. Nadav Brandes
5. Dan Ofer
6. Michal Linial
7. Jian Peng
8. Yufeng Su

In Representation & Self-Supervised Learning

1. Pieter Abbeel
2. Sergey Levine
3. Jian Peng
4. Yan Duan
5. Xi Chen
6. Chelsea Finn
7. Kevin Kaichuang Yang
8. Xinqiang Ding

In Computational Biology & Bioinformatics

1. Kevin Kaichuang Yang
2. Zachary Wu
3. Frances H. Arnold
4. Amy X. Lu
5. Jian Peng
6. Yan Duan
7. Yun S. Song
8. Huimin Zhao

Citations in the corpus

• Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences (2019, neutral): “In contrast to models that are based on recurrent or convolutional neural networks, the Transformer makes no assumptions on the ordering of the input and instead uses position embeddings.”
• Evolutionary context-integrated deep sequence modeling for protein engineering (2020, methodology): “Yang et al ​[46] proposed to use a Doc2Vec model ​[47] ​, pre-trained on ~500k UniProt sequences, to map an arbitrary sequence to a 64-dimensional real-valued vector.”
• Evaluating Protein Transfer Learning with TAPE (2019, background): “Other self-supervised work such as Alley et al. [12] and Yang et al. [33] report protein engineering results, but on different tasks and datasets.”
• Deciphering protein evolution and fitness landscapes with latent space models (2019, methodology): “58 is a generic doc2vec embedding method, which is learned by pooling sequences from many protein families together and viewing all protein sequences equally.”
• The language of proteins: NLP, machine learning & protein sequences (2021, methodology): “Despite its simplicity, BoW is an effective method for many protein-related tasks [42,113,79,66,64,69,86].”
• Self-Supervised Contrastive Learning of Protein Representations By Mutual Information Maximization (2020, methodology): “Though recent works have successfully demonstrated the ability to capture properties such as fluorescence, pair-wise contact, phylogenetics, structure, and subcellular localization, these works mostly use methods designed for natural language processing (NLP) [41, 31, 32, 3, 16, 11, 4, 24, 12].”
• Machine-learning-guided directed evolution for protein engineering (2018, background): “For small datasets, one-hot encodings offer superior performance to general sets of protein properties 73 , although careful feature selection informed by domain knowledge may yield more accurate predictions.”