The Use of ChatGPT in the World of BioInformatics and Cancer Research and Development of BioGPT by MIT
Curator: Stephen J. Williams, Ph.D.
Chatbots are being used in multiple interdisciplanary areas in research and medicine and so it was a natural progression to incorporate artificial intelligence (AI), natural language processing (NLP), and chatbot technology like ChatGPT into bioinformatic analysis.
“In domains like informatics, management and marketing, media and communication science, languages and philosophy, psychology and sociology, engineering, design, and human-computer interaction, the fast expanding body of chatbot study is clearly interdisciplinary.
The field of bioinformatics is a natural fit for incorporting this technology. The curated information below shows some use examples of this technology in bioinformatics related to cancer research.
The text in this reference source was actually generated by ChatGPT (https://chat.openai.com/chat)
Source: https://omicstutorials.com/using-chatgpt-in-bioinformatics-and-biomedical-research/
How ChatGPT can be used in bioinformatics research?
ChatGPT is a variant of the GPT (Generative Pre-trained Transformer) language model that was designed to generate human-like text in a conversational setting. It is not directly related to bioinformatics, which is the field of study that deals with the storage, retrieval, and analysis of biological data, particularly in the context of genetics and genomics. However, ChatGPT could potentially be used in a bioinformatics context as a tool for generating natural language descriptions of genetic or genomic data, or for generating responses to queries about such data.
For example, a researcher could use ChatGPT to generate descriptions of genetic variants or gene expression patterns based on input data. This could be useful for generating summaries of results for reports or papers, or for generating explanations of complex concepts for non-expert audiences. ChatGPT could also be used to generate responses to queries about genetic or genomic data, such as “What is the function of gene X?” or “What are the potential consequences of variant Y?”
It’s worth noting that ChatGPT is just one of many tools and techniques that can be used in bioinformatics, and it is unlikely to be the most suitable or effective option in every situation. There are many specialized tools and resources available for working with biological data, and it is often necessary to use a combination of these tools to fully analyze and understand the data.
Generating descriptions of genetic or genomic data: ChatGPT could be used to generate natural language descriptions of genetic or genomic data based on input data. For example, suppose a researcher has a dataset containing information about gene expression levels in different tissues. The researcher could use ChatGPT to generate a description of the data, such as:
“Gene X is highly expressed in the liver and kidney, with moderate expression in the brain and heart. Gene Y, on the other hand, shows low expression in all tissues except for the lung, where it is highly expressed.”
Thereby ChatGPT, at its simplest level, could be used to ask general questions like “What is the function of gene product X?” and a ChatGPT could give a reasonable response without the scientist having to browse through even highly curated databases lie GeneCards or UniProt or GenBank. Or even “What are potential interactors of Gene X, validated by yeast two hybrid?” without even going to the curated InterActome databases or using expensive software like Genie.
Summarizing results: ChatGPT could be used to generate summaries of results from genetic or genomic studies. For example, a researcher might use ChatGPT to generate a summary of a study that found a association between a particular genetic variant and a particular disease. The summary might look something like this:
“Our study found that individuals with the variant form of gene X are more likely to develop disease Y. Further analysis revealed that this variant is associated with changes in gene expression that may contribute to the development of the disease.”
It’s worth noting that ChatGPT is just one tool that could potentially be used in these types of applications, and it is likely to be most effective when used in combination with other bioinformatics tools and resources. For example, a researcher might use ChatGPT to generate a summary of results, but would also need to use other tools to analyze the data and confirm the findings.
ChatGPT is a variant of the GPT (Generative Pre-training Transformer) language model that is designed for open-domain conversation. It is not specifically designed for generating descriptions of genetic variants or gene expression patterns, but it can potentially be used for this purpose if you provide it with a sufficient amount of relevant training data and fine-tune it appropriately.
To use ChatGPT to generate descriptions of genetic variants or gene expression patterns, you would first need to obtain a large dataset of examples of descriptions of genetic variants or gene expression patterns. You could use this dataset to fine-tune the ChatGPT model on the task of generating descriptions of genetic variants or gene expression patterns.
Here’s an example of how you might use ChatGPT to generate a description of a genetic variant:
First, you would need to pre-process your dataset of descriptions of genetic variants to prepare it for use with ChatGPT. This might involve splitting the descriptions into individual sentences or phrases, and encoding them using a suitable natural language processing (NLP) library or tool.
Next, you would need to fine-tune the ChatGPT model on the task of generating descriptions of genetic variants. This could involve using a tool like Hugging Face’s Transformers library to load the ChatGPT model and your pre-processed dataset, and then training the model on the task of generating descriptions of genetic variants using an appropriate optimization algorithm.
Once the model has been fine-tuned, you can use it to generate descriptions of genetic variants by providing it with a prompt or seed text and asking it to generate a response. For example, you might provide the model with the prompt “Generate a description of a genetic variant associated with increased risk of breast cancer,” and ask it to generate a response. The model should then generate a description of a genetic variant that is associated with increased risk of breast cancer.
It’s worth noting that generating high-quality descriptions of genetic variants or gene expression patterns is a challenging task, and it may be difficult to achieve good results using a language model like ChatGPT without a large amount of relevant training data and careful fine-tuning.
To train a language model like chatGPT to extract information about specific genes or diseases from research papers, you would need to follow these steps:
Gather a large dataset of research papers that contain information about the specific genes or diseases you are interested in. This dataset should be diverse and representative of the types of papers you want the model to be able to extract information from.
Preprocess the text data in the research papers by tokenizing the text and creating a vocabulary. You may also want to consider lemmatizing or stemming the text to reduce the dimensionality of the dataset.
Train the language model on the preprocessed text data. You may want to fine-tune a pre-trained model such as chatGPT on your specific dataset, or you can train a new model from scratch.
ChatGPT could also be useful for sequence analysis
A few examples of sequence analysis a ChatGPT could be useful include:
- Protein structure
- Identifying functional regions of a protein
- Predicting protein-protein interactions
- Identifying protein homologs
- Generating Protein alignments
All this could be done without having access to UNIX servers or proprietary software or knowing GCG coding
ChatGPT in biomedical research
There are several potential ways that ChatGPT or other natural language processing (NLP) models could be applied in biomedical research:
Text summarization: ChatGPT or other NLP models could be used to summarize large amounts of text, such as research papers or clinical notes, in order to extract key information and insights more quickly.
Data extraction: ChatGPT or other NLP models could be used to extract structured data from unstructured text sources, such as research papers or clinical notes. For example, the model could be trained to extract information about specific genes or diseases from research papers, and then used to create a database of this information for further analysis.
Literature review: ChatGPT or other NLP models could be used to assist with literature review tasks, such as identifying relevant papers, extracting key information from papers, or summarizing the main findings of a group of papers.
Predictive modeling: ChatGPT or other NLP models could be used to build predictive models based on large amounts of text data, such as electronic health records or research papers. For example, the model could be trained to predict the likelihood of a patient developing a particular disease based on their medical history and other factors.
It’s worth noting that while NLP models like ChatGPT have the potential to be useful tools in biomedical research, they are only as good as the data they are trained on, and it is important to carefully evaluate the quality and reliability of any results generated by these models.
ChatGPT in text mining of biomedical data
ChatGPT could potentially be used for text mining in the biomedical field in a number of ways. Here are a few examples:
Extracting information from scientific papers: ChatGPT could be trained on a large dataset of scientific papers in the biomedical field, and then used to extract specific pieces of information from these papers, such as the names of compounds, their structures, and their potential uses.
Generating summaries of scientific papers: ChatGPT could be used to generate concise summaries of scientific papers in the biomedical field, highlighting the main findings and implications of the research.
Identifying trends and patterns in scientific literature: ChatGPT could be used to analyze large datasets of scientific papers in the biomedical field and identify trends and patterns in the data, such as emerging areas of research or common themes among different papers.
Generating questions for further research: ChatGPT could be used to suggest questions for further research in the biomedical field based on existing scientific literature, by identifying gaps in current knowledge or areas where further investigation is needed.
Generating hypotheses for scientific experiments: ChatGPT could be used to generate hypotheses for scientific experiments in the biomedical field based on existing scientific literature and data, by identifying potential relationships or associations that could be tested in future research.
PLEASE WATCH VIDEO
In this video, a bioinformatician describes the ways he uses ChatGPT to increase his productivity in writing bioinformatic code and conducting bioinformatic analyses.
He describes a series of uses of ChatGPT in his day to day work as a bioinformatian:
- Using ChatGPT as a search engine: He finds more useful and relevant search results than a standard Google or Yahoo search. This saves time as one does not have to pour through multiple pages to find information. However, a caveat is ChatGPT does NOT return sources, as highlighted in previous postings on this page. This feature of ChatGPT is probably why Microsoft bought OpenAI in order to incorporate ChatGPT in their Bing search engine, as well as Office Suite programs
- ChatGPT to help with coding projects: Bioinformaticians will spend multiple hours searching for and altering open access available code in order to run certain function like determining the G/C content of DNA (although there are many UNIX based code that has already been established for these purposes). One can use ChatGPT to find such a code and then assist in debugging that code for any flaws
- ChatGPT to document and add coding comments: When writing code it is useful to add comments periodically to assist other users to determine how the code works and also how the program flow works as well, including returned variables.
One of the comments was interesting and directed one to use BIOGPT instead of ChatGPT
@tzvi7989
0:54 oh dear. You cannot use chatgpt like that in Bioinformatics as it is rn without double checking the info from it. You should be using biogpt instead for paper summarisation. ChatGPT goes for human-like responses over precise information recal. It is quite good for debugging though and automating boring awkward scripts
So what is BIOGPT?
BioGPT https://github.com/microsoft/BioGPT
The BioGPT model was proposed in BioGPT: generative pre-trained transformer for biomedical text generation and mining by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu. BioGPT is a domain-specific generative pre-trained Transformer language model for biomedical text generation and mining. BioGPT follows the Transformer language model backbone, and is pre-trained on 15M PubMed abstracts from scratch.
The abstract from the paper is the following:
Pre-trained language models have attracted increasing attention in the biomedical domain, inspired by their great success in the general natural language domain. Among the two main branches of pre-trained language models in the general language domain, i.e. BERT (and its variants) and GPT (and its variants), the first one has been extensively studied in the biomedical domain, such as BioBERT and PubMedBERT. While they have achieved great success on a variety of discriminative downstream biomedical tasks, the lack of generation ability constrains their application scope. In this paper, we propose BioGPT, a domain-specific generative Transformer language model pre-trained on large-scale biomedical literature. We evaluate BioGPT on six biomedical natural language processing tasks and demonstrate that our model outperforms previous models on most tasks. Especially, we get 44.98%, 38.42% and 40.76% F1 score on BC5CDR, KD-DTI and DDI end-to-end relation extraction tasks, respectively, and 78.2% accuracy on PubMedQA, creating a new record. Our case study on text generation further demonstrates the advantage of BioGPT on biomedical literature to generate fluent descriptions for biomedical terms.
Tips:
- BioGPT is a model with absolute position embeddings so it’s usually advised to pad the inputs on the right rather than the left.
- BioGPT was trained with a causal language modeling (CLM) objective and is therefore powerful at predicting the next token in a sequence. Leveraging this feature allows BioGPT to generate syntactically coherent text as it can be observed in the run_generation.py example script.
- The model can take the past_key_values (for PyTorch) as input, which is the previously computed key/value attention pairs. Using this (past_key_values or past) value prevents the model from re-computing pre-computed values in the context of text generation. For PyTorch, see past_key_values argument of the BioGptForCausalLM.forward() method for more information on its usage.
This model was contributed by kamalkraj. The original code can be found here.
This repository contains the implementation of BioGPT: Generative Pre-trained Transformer for Biomedical Text Generation and Mining, by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu. BioGPT is a github which is being developed by MIT in collaboration with Microsoft. It is based on Python.
License
BioGPT is MIT-licensed. The license applies to the pre-trained models as well.
Contributing
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.
When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
As of right now this does not seem Open Access, however a sign up is required!
We provide our pre-trained BioGPT model checkpoints along with fine-tuned checkpoints for downstream tasks, available both through URL download as well as through the Hugging Face 🤗 Hub.
| Model | Description | URL | 🤗 Hub |
| BioGPT | Pre-trained BioGPT model checkpoint | link | link |
| BioGPT-Large | Pre-trained BioGPT-Large model checkpoint | link | link |
| BioGPT-QA-PubMedQA-BioGPT | Fine-tuned BioGPT for question answering task on PubMedQA | link | |
| BioGPT-QA-PubMedQA-BioGPT-Large | Fine-tuned BioGPT-Large for question answering task on PubMedQA | link | |
| BioGPT-RE-BC5CDR | Fine-tuned BioGPT for relation extraction task on BC5CDR | link | |
| BioGPT-RE-DDI | Fine-tuned BioGPT for relation extraction task on DDI | link | |
| BioGPT-RE-DTI | Fine-tuned BioGPT for relation extraction task on KD-DTI | link | |
| BioGPT-DC-HoC | Fine-tuned BioGPT for document classification task on HoC | link |
So far as to date (6/12/2023) there are three papers referencing BioGPT on PubMed (abstracts and links below)
BioGPT: generative pre-trained transformer for biomedical text generation and mining.
Luo R, Sun L, Xia Y, Qin T, Zhang S, Poon H, Liu TY.Brief Bioinform. 2022 Nov 19;23(6):bbac409. doi: 10.1093/bib/bbac409.PMID: 36156661
We evaluate BioGPT on six biomedical natural language processing tasks and demonstrate that our model outperforms previous models on most tasks. …Our case study on text generation further demonstrates the advantage of BioGPT on biomedical literature to generate fl …
Here they show the advantages of BioGPT over BioBERT and ChatGPT in an example of text extraction and question answer
Table 7. Example of generated text
Input Model Text
Bicalutamide GPT-2 Bicalutamide, a synthetic bicalutamide derivative, was found to increase the cellular proliferation
of C. elegans cells. This effect was not observed with the other synthetic bicalutamide derivatives. This is in line with our previous work which showed that bicalutamide increases the proliferation of C. elegans cells but not in the presence of synthetic bicalutamide.
BioGPT Bicalutamide (Casodex) is an androgen receptor (AR) antagonist approved for the treatment of metastatic castration-resistant prostate cancer (mCRPC) in patients who have progressed on or are ineligible for docetaxel chemotherapy, as well as for the treatment of early-stage prostate cancer in men who have not progressed on or are ineligible for docetaxel chemotherapy, as well as for the treatment of metastatic castration-sensitive prostate cancer (mCSPC) in men who have not progressed on or are ineligible for docetaxel chemotherapy.
Janus kinase 3 (JAK-3) GPT-2 Janus kinase 3 (JAK-3) mediates the conversion of glucose to glycogen in skeletal muscle. The
increase in glucose uptake by skeletal muscle is believed to be associated with a reduction in muscle protein breakdown and with the release of ketone bodies, which contribute to the metabolism of glucose.
BioGPT Janus kinase 3 (JAK-3) is a member of the Janus kinase (JAK) family of non-receptor tyrosine kinases and plays an important role in the regulation of cell proliferation, differentiation, survival, migration and angiogenesis.
Apricitabine GPT-2 Apricitabine, duloxetine and citalopram) were investigated in patients with a history of
treatment-resistant depression (n = 10). We found that treatment with the three drugs significantly decreased depressive symptoms, but not depressive symptoms over the next 12 weeks. We conclude that antidepressant drugs that reduce depressive symptoms and reduce depressive symptoms over time may be beneficial in patients with recurrent depression.
BioGPT Apricitabine is an oral prodrug of 5-aza-2 ’-deoxycytidine (5-aza-CdR), a DNA methyltransferase (DNMT) inhibitor, which has been approved by the US Food and Drug Administration (FDA) for the treatment of myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML) in combination with low-dose cytarabine (Ara-C) and granulocyte colony-stimulating factor (G-CSF) for patients with intermediate-2 or high-risk MDS or AML.
Xylazine GPT-2 Xylazine, the ‘active ingredient’ in ‘bath salts’ (see above) has been linked to numerous deaths,
including those of former University of California, Berkeley, professor Andrew Anglin, who posted
a video of himself having a bath filled with it.
BioGPT Xylazine is an alpha 2-adrenoceptor agonist which has been used as a sedative and analgesic in veterinary medicine for many years, but its effects on the cardiovascular system have not been extensively studied in the dog, and its effects on the central nervous system (CNS) have not been well characterized in the dog, despite the fact that xylazine has been widely used as a sedative and analgesic in veterinary medicine for more than 30 years.
Psoralen GPT-2 Psoralen, the city of the spiritualist, and had been given to him by God, who wished to show him
the way of salvation, and to draw him back to the true life of Christ, and to bring him into communion with his fellow-man.
BioGPT Psoralen photochemotherapy (PUVA) is a well-established treatment for psoriasis, but its use is limited by the risk of skin cancer, particularly squamous cell carcinoma (SCC) of the head and neck (H & N), which is the most common site of skin cancer in the United States (US)
CP-673451 GPT-2 CP-673451) and 4-5-6-3-1-2-3-4-6-7-8-9-10-11-12-13-14-15-16-17-18-19-20-21-22-23-24-25-26-27-
28-29-30-31-32-33-34-35-36-37-38-39-40-41-42-43-44-45-46-47-48-49-50…
BioGPT CP-673451 is a potent, selective, and orally active inhibitor of human neutrophil elastase (HNE) and human cathepsin G (CatG) with in vitro and in vivo anti-inflammatory activity in a variety of animal models of inflammation and in a model of acute lung injury (ALI) in the rat induced by intratracheal instillation of lipopolysaccharide (LPS) and tumor necrosis factor-alpha (TNF-alpha), a model of acute lung injury (ALI) in which neutrophils play an important role.
BIIB-021 GPT-2 BIIB-021A-2-1.pdf. Wright, C. (2002). ‘The War on Terror: A Legal Theory of Terrorist Attacks,’ in
L.E.R. Wright, S.R. Warrick and J.H. Rothman (Eds.), The War on Terror: Essays in Honor of Michael T. Klare (New York: Oxford University Press), 143-57.
BioGPT BIIB-021 is a novel, orally active, non-peptide bradykinin B2 receptor antagonist with potent and long-lasting anti-inflammatory activity in animal models of acute and chronic inflammation and in a rat model of adjuvant-induced arthritis (AIA), an animal model of rheumatoid arthritis (RA) and in a rat model of collagen-induced arthritis (CIA), an animal model of collagen-induced arthritis (CIA), in which arthritis is induced by immunization with bovine type II collagen (CII).
Note how BioGPT is more descriptive and accurate!
Huang L, Lin J, Li X, Song L, Zheng Z, Wong KC.Brief Bioinform. 2022 Jan 17;23(1):bbab451. doi: 10.1093/bib/bbab451.PMID: 34791012
The rapid growth in literature accumulates diverse and yet comprehensive biomedical knowledge hidden to be mined such as drug interactions. However, it is difficult to extract the heterogeneous knowledge to retrieve or even discover the latest and novel knowledge in an efficient manner. To address such a problem, we propose EGFI for extracting and consolidating drug interactions from large-scale medical literature text data. Specifically, EGFI consists of two parts: classification and generation. In the classification part, EGFI encompasses the language model BioBERT which has been comprehensively pretrained on biomedical corpus. In particular, we propose the multihead self-attention mechanism and packed BiGRU to fuse multiple semantic information for rigorous context modeling. In the generation part, EGFI utilizes another pretrained language model BioGPT-2 where the generation sentences are selected based on filtering rules.
Results: We evaluated the classification part on ‘DDIs 2013’ dataset and ‘DTIs’ dataset, achieving the F1 scores of 0.842 and 0.720 respectively. Moreover, we applied the classification part to distinguish high-quality generated sentences and verified with the existing growth truth to confirm the filtered sentences. The generated sentences that are not recorded in DrugBank and DDIs 2013 dataset demonstrated the potential of EGFI to identify novel drug relationships.
Availability: Source code are publicly available at https://github.com/Layne-Huang/EGFI.
Jin Q, Yang Y, Chen Q, Lu Z.ArXiv. 2023 May 16:arXiv:2304.09667v3. Preprint.PMID: 37131884 Free PMC article.
While large language models (LLMs) have been successfully applied to various tasks, they still face challenges with hallucinations. Augmenting LLMs with domain-specific tools such as database utilities can facilitate easier and more precise access to specialized knowledge. In this paper, we present GeneGPT, a novel method for teaching LLMs to use the Web APIs of the National Center for Biotechnology Information (NCBI) for answering genomics questions. Specifically, we prompt Codex to solve the GeneTuring tests with NCBI Web APIs by in-context learning and an augmented decoding algorithm that can detect and execute API calls. Experimental results show that GeneGPT achieves state-of-the-art performance on eight tasks in the GeneTuring benchmark with an average score of 0.83, largely surpassing retrieval-augmented LLMs such as the new Bing (0.44), biomedical LLMs such as BioMedLM (0.08) and BioGPT (0.04), as well as GPT-3 (0.16) and ChatGPT (0.12). Our further analyses suggest that: (1) API demonstrations have good cross-task generalizability and are more useful than documentations for in-context learning; (2) GeneGPT can generalize to longer chains of API calls and answer multi-hop questions in GeneHop, a novel dataset introduced in this work; (3) Different types of errors are enriched in different tasks, providing valuable insights for future improvements.
PLEASE WATCH THE FOLLOWING VIDEOS ON BIOGPT
This one entitled
Microsoft’s BioGPT Shows Promise as the Best Biomedical NLP
gives a good general description of this new MIT/Microsoft project and its usefullness in scanning 15 million articles on PubMed while returning ChatGPT like answers.
Please note one of the comments which is VERY IMPORTANT
@rufus9322
bioGPT is difficult for non-developers to use, and Microsoft researchers seem to default that all users are proficient in Python and ML.
Much like Microsoft Azure it seems this BioGPT is meant for developers who have advanced programming skill. Seems odd then to be paying programmers multiK salaries when one or two Key Opinion Leaders from the medical field might suffice but I would be sure Microsoft will figure this out.
ALSO VIEW VIDEO
This is a talk from Microsoft on BioGPT
Other Relevant Articles on Natural Language Processing in BioInformatics, Healthcare and ChatGPT for Medicine on this Open Access Scientific Journal Include
Medicine with GPT-4 & ChatGPT
Explanation on “Results of Medical Text Analysis with Natural Language Processing (NLP) presented in LPBI Group’s NEW GENRE Edition: NLP” on Genomics content, standalone volume in Series B and NLP on Cancer content as Part B New Genre Volume 1 in Series C
From High-Throughput Assay to Systems Biology: New Tools for Drug Discovery
20 articles in Natural Language Processing
142 articles in BioIT: BioInformatics
111 articles in BioIT: BioInformatics, NGS, Clinical & Translational, Pharmaceutical R&D Informatics, Clinical Genomics, Cancer Informatics
2012pharmaceutical
sjwilliamspa