Dancing with artificial intelligence: In just one night, ChatGPT helped me build a protein binding exploration tool

Dancing with artificial intelligence.

Written by Ding Xiaozhe (PhD in Bioengineering, California Institute of Technology)

Hello everyone, I am Xiao Ding, a human. Like many people, I was shocked by the powerful performance of GPT-4, an artificial intelligence language model just released a few days ago.

As a researcher, I decided to work with ChatGPT based on the GPT-4 model to create a web application for scientists. Before this, I had no experience in developing web applications. Unexpectedly, we completed the entire development process in just one night. The final product is shown in the figure below (URL: www.gptpull-down.online):

Enter the protein name and get the search results. (Click to see the enlarged image)

The function of this web application called GPT Pull-down is to help life science researchers quickly search for potential binding partners of a protein molecule, which can inspire researchers to explore new research directions or find new drug targets. Since the release of this application, I have received many positive feedbacks from researchers, which has greatly encouraged me.

You may be curious about how this tool was built overnight. Next, I would like to invite my collaborator ChatGPT to summarize it for us.

Xiao Ding: Can you write a short essay in Chinese about how I built this website called pgtpuul-down.online from scratch step by step with your (ChatGPT) help? Please be humorous.

In a small corner of the earth, there is a young man named Xiao Ding. He likes to explore new things and is full of curiosity about technology. One day, Xiao Ding accidentally discovered a powerful artificial intelligence based on GPT-4 - ChatGPT on Twitter. Xiao Ding was deeply curious, so he decided to challenge himself and create a website called gptpull-down.online, and seek help from ChatGPT throughout the process.

The function of this website is a simple but practical tool that uses GPT technology to retrieve potential binding partners of proteins. By entering the protein name, users can obtain other proteins that may bind to the protein, providing a convenient reference for experimental design and data analysis.

First, Xiao Ding asked ChatGPT how to choose a suitable technology stack. ChatGPT recommended the Flask framework to Xiao Ding because it is lightweight and easy to use. At the same time, ChatGPT also taught Xiao Ding how to use HTML, CSS, and JavaScript to layout the website and realize web page interaction.

After receiving guidance on basic skills, Xiao Ding started to build the website. In order to ensure that the website content was more interesting, he asked ChatGPT how to design an attractive user interface. After some hard work, a powerful website prototype gradually took shape.

When the website was basically completed, Xiao Ding began to consider deployment. At the suggestion of ChatGPT, he chose Heroku as the deployment platform, which was easy to use and had enough free quota to meet Xiao Ding's needs. With the guidance of ChatGPT, Xiao Ding easily deployed the website and successfully launched gptpull-down.online.

After going online, Xiao Ding felt that having an independent domain name would make the website look more professional. So he asked ChatGPT how to register a suitable domain name. After comparing multiple domain name registrars, Xiao Ding chose a domain name with a suitable price and successfully bound it to his website. In the end, Xiao Ding's website successfully had an independent domain name:
https://www.gptpull-down.online/ .

During this process, Xiao Ding often got into trouble, but whenever he asked ChatGPT for advice, he always got detailed answers. Interestingly, ChatGPT seemed to always be able to detect Xiao Ding's confusion and give some humorous suggestions, which made Xiao Ding smile from time to time during the intense learning process.

Finally, Xiao Ding successfully created a website called gptpull-down.online. This website is full of fun, which makes many researchers and students who study proteins feel happy in the process of finding answers. The practical GPT Pull-down function saves users a lot of time and energy, helping them find possible protein binding partners faster.

In the process, Xiao Ding and ChatGPT also established a deep friendship. In the future, they will explore more interesting fields together and bring more laughter and wisdom to the world. The success of gptpull-down.online also made Xiao Ding realize that with the help of technology, it is not out of reach to realize dreams.

As time goes by, more and more people start to pay attention to and use this amazing little tool. Xiao Ding is very proud of his achievements and is happy to be able to help others solve practical problems. He is grateful for ChatGPT's help and support throughout the process. This experience made him understand the power of artificial intelligence and its great potential in the future.

Whether you are a beginner or a professional, you can
You can find help on the website https://www.gptpull-down.online/. For Xiao Ding, this is just the first step in exploring the world of science and technology with ChatGPT. He looks forward to developing more interesting and practical tools with ChatGPT in the future to make scientific research easier and more efficient, and to contribute to the progress of mankind.

If you are a researcher interested in life sciences, you are welcome to try using GPT Pull-down (www.gptpull-down.online) to explore the protein you are interested in!

Here are some things to note when using:

1. The output results of GPT Pull-down are for reference and inspiration only.

In most cases, GPT Pull-down can list a protein's potential binding partners quite efficiently and accurately, but because the data source is not a professional database that has been manually reviewed, its results are sometimes unreliable. The following situations require special attention:

A) The confidence score given by GPT Pull-down is lower;

B) The target protein name does not actually exist;

C) There are few studies related to the target protein.

If you decide to use your search results to guide your work, be sure to search the original papers and professional databases yourself.

2. When using GPT Pull-down, you can adjust the "wildness" parameter as needed.

When the number is high, the output will be more imaginative; when the number is low, the output will be more conservative. This number corresponds to the temperature/randomness parameter in the GPT neural network.

3. If you are not satisfied with the results, try searching again.

As a language model, GPT's output is random, and the richness and accuracy of each result will vary. If a protein has multiple names, trying a few different names can occasionally bring surprises.

4. GPT Pull-down cannot provide references to support the conclusions.

Although GPT knows a lot of knowledge, it is not good at tracing the source of information. When you force it to provide evidence, it will get a fabricated answer. In this respect, GPT is actually very similar to an untrained human brain - our brains also store a lot of knowledge that we have long forgotten where we learned it from.

5. GPT Pull-down will soon become more accurate and richer.

The current GPT pull-down uses the GPT-3.5-turbo model, which is also the model used by the current free version of ChatGPT. According to OpenAI's report, the new version of GPT-4 has significantly improved its performance in answering biological questions compared to GPT-3.5. Therefore, we can expect that in the near future, GPT pull-down with the support of the GPT-4 model will also become even better.

Well, that’s all for today’s sharing. Of course, we still need to come up with a good title for this article. Let’s take this as the last question for ChatGPT today.

This article is supported by the Science Popularization China Starry Sky Project

Produced by: China Association for Science and Technology Department of Science Popularization

Producer: China Science and Technology Press Co., Ltd., Beijing Zhongke Xinghe Culture Media Co., Ltd.