There seems to be lots of research on turning protein sequences to embeddings (numeric vectors). I have a bunch of amino acid sequences that are length 30-50. Any suggestions on tools that do the embedding easily?
\nThanks a lot!
", "messages": [ { "id": "2285743", "postDate": "06/03/2023 00:28:18", "content": "There seems to be lots of research on turning protein sequences to embeddings (numeric vectors). I have a bunch of amino acid sequences that are length 30-50. Any suggestions on tools that do the embedding easily?
\nThanks a lot!
", "rawMarkdown": "There seems to be lots of research on turning protein sequences to embeddings (numeric vectors). I have a bunch of amino acid sequences that are length 30-50. Any suggestions on tools that do the embedding easily?\n\nThanks a lot!", "votes": null }, { "id": "2288898", "postDate": "06/05/2023 18:40:16", "content": "In general (not precisely for amino acid sequences), transform text sequences into vector embeddings is a huge branch of research in Natural Langage Processing (NLP). In general it is based on some mathematical tools that aims to compute text similarity between two text sequences. To start easily I recommend you this link https://www.analyticsvidhya.com/blog/2021/06/part-5-step-by-step-guide-to-master-nlp-text-vectorization-approaches/ that talks about co-occurence matrix that simply count in a matrix form the number of tokens that append in a sequence.
\nObviously there are more complex tools that give importance to the order of tokens and their context (Continuous Bag Of Words CBOW, SkipGram). The general framework is to predict context given the word or predict the word given the context. This will lead to vector representation. I recommend this reading https://fasttext.cc/docs/en/unsupervised-tutorial.html#:~:text=Advanced%20readers%3A%20skipgram%20versus%20cbow&text=The%20skipgram%20model%20learns%20to,window%20around%20the%20target%20word. that talks about FastText, a model trained on Wikipedia pages for vector embeddings.
\nDont really know for Biomedial precisely, but there are T5, ESM2, ProtBERT that are langage models trained specifically on protein sequences.
", "rawMarkdown": "In general (not precisely for amino acid sequences), transform text sequences into vector embeddings is a huge branch of research in Natural Langage Processing (NLP). In general it is based on some mathematical tools that aims to compute text similarity between two text sequences. To start easily I recommend you this link https://www.analyticsvidhya.com/blog/2021/06/part-5-step-by-step-guide-to-master-nlp-text-vectorization-approaches/ that talks about co-occurence matrix that simply count in a matrix form the number of tokens that append in a sequence. \n\nObviously there are more complex tools that give importance to the order of tokens and their context (Continuous Bag Of Words CBOW, SkipGram). The general framework is to predict context given the word or predict the word given the context. This will lead to vector representation. I recommend this reading https://fasttext.cc/docs/en/unsupervised-tutorial.html#:~:text=Advanced%20readers%3A%20skipgram%20versus%20cbow&text=The%20skipgram%20model%20learns%20to,window%20around%20the%20target%20word. that talks about FastText, a model trained on Wikipedia pages for vector embeddings. \n\nDont really know for Biomedial precisely, but there are T5, ESM2, ProtBERT that are langage models trained specifically on protein sequences.", "votes": null }, { "id": "2315277", "postDate": "06/24/2023 02:35:17", "content": "one-hot embedding for such small seq length ? but we have 35K length in the data set..
", "rawMarkdown": "one-hot embedding for such small seq length ? but we have 35K length in the data set..", "votes": null } ], "comments": [ { "id": 2288898, "author_name": "henriupton", "author_url": "", "post_date": "06/05/2023 18:40:16", "content": "In general (not precisely for amino acid sequences), transform text sequences into vector embeddings is a huge branch of research in Natural Langage Processing (NLP). In general it is based on some mathematical tools that aims to compute text similarity between two text sequences. To start easily I recommend you this link https://www.analyticsvidhya.com/blog/2021/06/part-5-step-by-step-guide-to-master-nlp-text-vectorization-approaches/ that talks about co-occurence matrix that simply count in a matrix form the number of tokens that append in a sequence.
\nObviously there are more complex tools that give importance to the order of tokens and their context (Continuous Bag Of Words CBOW, SkipGram). The general framework is to predict context given the word or predict the word given the context. This will lead to vector representation. I recommend this reading https://fasttext.cc/docs/en/unsupervised-tutorial.html#:~:text=Advanced%20readers%3A%20skipgram%20versus%20cbow&text=The%20skipgram%20model%20learns%20to,window%20around%20the%20target%20word. that talks about FastText, a model trained on Wikipedia pages for vector embeddings.
\nDont really know for Biomedial precisely, but there are T5, ESM2, ProtBERT that are langage models trained specifically on protein sequences.
", "votes": null, "replies": [] }, { "id": 2315277, "author_name": "sunilkrishnan", "author_url": "", "post_date": "06/24/2023 02:35:17", "content": "one-hot embedding for such small seq length ? but we have 35K length in the data set..
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2285743": "There seems to be lots of research on turning protein sequences to embeddings (numeric vectors). I have a bunch of amino acid sequences that are length 30-50. Any suggestions on tools that do the embedding easily?\n\nThanks a lot!", "2288898": "In general (not precisely for amino acid sequences), transform text sequences into vector embeddings is a huge branch of research in Natural Langage Processing (NLP). In general it is based on some mathematical tools that aims to compute text similarity between two text sequences. To start easily I recommend you this link https://www.analyticsvidhya.com/blog/2021/06/part-5-step-by-step-guide-to-master-nlp-text-vectorization-approaches/ that talks about co-occurence matrix that simply count in a matrix form the number of tokens that append in a sequence. \n\nObviously there are more complex tools that give importance to the order of tokens and their context (Continuous Bag Of Words CBOW, SkipGram). The general framework is to predict context given the word or predict the word given the context. This will lead to vector representation. I recommend this reading https://fasttext.cc/docs/en/unsupervised-tutorial.html#:~:text=Advanced%20readers%3A%20skipgram%20versus%20cbow&text=The%20skipgram%20model%20learns%20to,window%20around%20the%20target%20word. that talks about FastText, a model trained on Wikipedia pages for vector embeddings. \n\nDont really know for Biomedial precisely, but there are T5, ESM2, ProtBERT that are langage models trained specifically on protein sequences.", "2315277": "one-hot embedding for such small seq length ? but we have 35K length in the data set.." }, "source": "meta" }