Here is my approach to the CAFA6 challenge: predicting protein function from sequence. It is based on converting everything into numbers, using ESM2 for protein sequences, and an LLM for Gene Ontology terms.
gmdbioinformatics.substack.com/p/my-strateg...
#CAFA6 #ESM2 #deepLearning #bioinformatics
The winners of the Arc Virtual Cell Challenge have been announced!
Did a foundation model outperform regressions?
Did the leaderboard hacking matter in the end?
Here are my thoughts. gmdbioinformatics.substack.com/p/arc-virtua...
#ArcVirtualCellChallenge #VirtualCellChallenge
Proud to have contributed to this bioArXiv paper, presenting an Agentic System for Cell Type annotation in single-cell
open.substack.com/pub/gmdbioin...
🧬 🖥️
#singlecell #bioinformatics #ngs #cytetype
Proud to have contributed to this bioArXiv paper, presenting an Agentic System for Cell Type annotation in single-cell
open.substack.com/pub/gmdbioin...
#singlecell #bioinformatics #ngs #cytetype
My third article on the Arc Virtual Cell Challenge.
This time, we learn how to train our very first Virtual Cell Model.
gmdbioinformatics.substack.com/p/arc-virtua...
#ArcInstitute #VirtualCellChallenge #BiologicalModelling #FoundationModels #MachineLearning #ML
I'm participating in the Arc Virtual Cell Challenge. The objective is to predict the effect of gene perturbations in H1 cell lines.
Here is my diary - more posts to come soon!
gmdbioinformatics.substack.com/p/arc-virtua...
#bio #bioinformatics #arc #foundation_models
I'm participating in the Arc Virtual Cell Challenge. The objective is to predict the effect of gene perturbations in H1 cell lines.
Here is my diary - more posts to come soon!
gmdbioinformatics.substack.com/p/arc-virtua...
#foundation-models #arc @arcinstitute.org #ML
My 8yo son and I developed a maths game called Times Tables Animals, featuring cute bunnies, maths, and volcanoes. Here is our story.
dalloliogm.github.io/Times-Table-...
@kilocode.ai @ttrockstars.bsky.social #llm #parents-ai
I joined Kaggle about 15 years ago. However, I never had time to complete the registration until earlier this year. Am I the person who has taken the most time to reach the Expert level on Kaggle? Let's look at the numbers :-)
gmdbioinformatics.substack.com/p/am-i-the-s...
In my opinion, MLflow is the missing piece for ensuring reproducibility of analysis in Bioinformatics. However, not many people in the field are familiar with it. Here I provide an example based on Differential Expression. medium.com/@dalloliogm/...
#Bioinformatics
#mlflow🧬🖥️
Foundation Models for Biology are going to revolutionize Bioinformatics in 2025. I wrote a short article to explain what they are and how they can be used. #llm #bioinformatics
medium.com/@dalloliogm/...
These two courses on DeepLearning.ai complement each other. The first one is about pre-training a LLM, and the second about fine-tuning them.
- learn.deeplearning.ai/courses/pret...
- learn.deeplearning.ai/courses/fine...
#llm-challenge #ai-challenge
o1 is the "Wise" model from OpenAI, the one that "thinks before it speaks" and thinks with its virtual right hemisphere.
This course summarizes the differences between o1 and other models, and how to optimize costs and times.
#ai-challenge #llm-challenge
learn.deeplearning.ai/accomplishme...
Just finished this course on Understanding and Applying Text Embedding.
Short and useful materials to understand how text can be embedded in a vector space, based on concept similarity, and used for training new models. #ai-challenge #llm-challenge
learn.deeplearning.ai/accomplishme...
I've just finished catching up on the 5-Day GenAI course from Kaggle and Google (www.kaggle.com/learn-guide/...)
I recommend this course to anyone wishing to learn more about LLMs. The explanations are clear, from the basics to the most technical parts. The Kaggle notebooks show concrete examples.
2)
They also provide two podcasts where two speakers summarize the documents. The funny thing is, this podcast is completely generated by AI, and the two speakers are not real people!
www.youtube.com/watch?v=mQDl...
www.youtube.com/watch?v=F_hJ...
1) They provide two white papers on Transformer Architecture and Prompt Engineering. A good resource for anyone wishing to learn more about these topics.
www.kaggle.com/whitepaper-f...
www.kaggle.com/whitepaper-p...
Ahah grazie :-)
Mi piacerebbe espandere la parte sugli agenti. Avevo solo 15 minuti per presentare, e non sono riuscito ad entrare in dettagli.
I'm catching up on the 5-day-GenAI course from Kaggle.
The first day introduces the Transformer architecture and the most important prompting techniques.
www.kaggle.com/learn-guide/...
#llm-challenge #ai-challenge
LLMs are transforming drug discovery, but many outside the field don't know about it. Science communication is key to bridging this gap. 🧪🎄 Here is a short talk I gave at the 'Salute è Donna' conference in Lanciano, Italy. ✨ #AI #DrugDiscovery"
www.slideshare.net/slideshow/ap...
This course gives a nice intro to Diffusion Models, including some examples of code for training a small diffusion model and generating some images in a notebook. learn.deeplearning.ai/courses/diff...
Today I've completed this course from Kaggle, on Machine Learning explanability. It's a great course, including partial dependency plots, permutation importance, and SHAP values. #ai-challenge
llm-challenge www.kaggle.com/learn/machin...
Kaggle has a collection of excellent free courses on Data Science. Today I've completed the Intro to Deep Learning, which is a good refresher on the basis of TensorFlow and Keras to train neural networks. #ai-challenge #llm-challenge
www.kaggle.com/learn/intro-...
Most of these things can be achieved by copy&pasting from chatGPT, but in the course they use API calls to VertexAI, using text-bison-001, to better template it.
This course makes some good examples of pairing with a LLM:
- Improving existing code
- Simplify code
- Write Test Cases
- Make code more efficient
- Debug code
- Create documentation to reduce technical debt
Excellent course on Pair Programming with a Large Language Model!
I think this is the way of the future. Every programmer or data scientist should learn to use LLMs to improve their code.
learn.deeplearning.ai/accomplishme...
Key takeaways:
- they create a fantasy world using a hierarchical procedure, from world->kingdom->towns, using LLM queries and JSON to store data
- they use TogetherAI's APIs to serve models, and `gradio` as a textual UI
- content regulation and safety is done using LlamaGuard.
Another course on implementing applications using LLMs. In this case, they do something that any nerd should try at least once: creating a text-based role-play fantasy game.
learn.deeplearning.ai/accomplishme... #llm-challenge #ai-challenge
I've decided to start a personal challenge to upskill my LLM and AI/ML knowledge! Over the next month, until Christmas, I'll try to complete a course on LLM, transformers, and deep learning every day. Wish me luck! #llm-challenge #upskill #llm #ai-challenge
