Eric J Ma's Website

In this blog post, I share my journey of learning to script PyMOL directly from Jupyter notebooks, a skill I picked up with the help of GPT-4. I detail the process of installing PyMOL, setting up the environment, and scripting to process and visualize protein structures, specifically nanobody VHH structures. I also demonstrate how to plot these structures in a grid using matplotlib. But the more important lesson here is how quickly I was able to pick it up, thanks to GPT-4! Are you able to leverage LLMs as a skill-learning hack?

In this blog post, I do a deep dive into a fascinating paper on designing CRISPR-Cas sequences using machine learning. The authors develop a generative model to produce novel protein sequences, validated in the lab, aiming to circumvent intellectual property restrictions. They curate a vast dataset, the CRISPR-Cas Atlas, and employ various models and filters to ensure sequence viability. My review highlights the methodology, emphasizing the importance of filtering and the challenges of using 'magic numbers' without justification. How many sequences are enough to train a generative model, and what makes laboratory experiments faster? Curious to find out more?

In this blog post, I share discussion insights from a hands-off tutorial I led at ODSC East on setting up a successful data science team within a biotech research organization. We explored formulating a mission, identifying problem classes, articulating value, and addressing challenges. I used my experience at Moderna to illustrate points, emphasizing the unique aspects of biotech data science. Despite not covering all topics due to time constraints, the discussion was enlightening, highlighting the contrast between biotech and other industries. How can these insights apply to your organization's data science team?

In this blog post, I share insights from my talk at the BioIT World conference in 2024, focusing on how LLMs empower data scientists and the necessity of software development skills in data science. I discuss practical applications of LLMs, such as code completion, documentation, debugging, and learning new domains, highlighting their role in enhancing productivity and efficiency. LLMs not only automate mundane tasks but also facilitate rapid knowledge acquisition, proving to be invaluable tools for data science teams. How could LLMs transform your data science work?

In this blog post, I share my journey of creating my first distributable pre-commit hook, convert-to-webp, using the pre-commit framework. This hook automatically converts images to the .webp format before they're committed to a repository, ensuring optimized image storage. I detail the essential configuration files, the creation of a Typer CLI for the hook, and how to make the hook available for others by tagging versions and adding it to a project's .pre-commit-config.yaml file. Curious about how to streamline your codebase with automated checks? How might this improve your project's efficiency?

In this blog post, I share the latest updates to pyds-cli, including the use of cookiecutter templates for easy repo scaffolding and a new talks initializer for creating talk presentations using reveal-md. These updates simplify the CLI and offer a streamlined approach to project and talk setup, reflecting my commitment to promoting best practices among data scientists. With these tools, I aim to make it easier for data scientists to adopt standardized project structures. Curious about how these updates can enhance your workflow?

In this blog post, I share insights from my 7 years in the industry on how to enhance a data science team's software development skills, focusing on the necessity of tooling and practices that make it easy and normal to do the right thing: moving from notebook explorations to production-ready code. I also discuss the importance of community practices in fostering a culture of quality software development within data science teams. How can these strategies streamline your team's workflow and elevate their software development capabilities?

In this blog post, I share the latest updates of LlamaBot 0.4.0, highlighting the decoupling of document storage from text generation in QueryBot, the introduction of the ChatUIMixin for easy web UI integration, and the switch to LanceDB for its lightweight, SQLite-like handling of vectors. I also touch on enhancements to repo chat, making it simpler to launch web-based chatbots on repository content. If you're a llamabot user, I'd love to hear from you about how well it works for you!

In this blog post, I discuss about organizing and motivating a data science team within a biotech research setting, focusing on structuring team activities around key research entities and methodologies. I highlight the importance of aligning team members with projects that match their interests and professional goals, and suggest ways to foster leadership skills without formal management roles. How do we balance the technical and career aspirations of data scientists to maintain productivity and motivation?

In this blog post, I share my experience running the Mixtral 8x7b-Instruct model on my old Linux GPU tower. I used the 4-bit quantized model and was pleasantly surprised that it worked. I generated keywords for a paper on protein engineering and machine learning using the model, and the results were comparable to GPT-4. Although the model was slower than running mistral-7b, it was still functional on older hardware. Have you tried running large language models on older hardware? Read on to find out more about my experience.