Mohamed Elkerdawy

Chief Information Officer & Computational Biologist - @Proteinea / Socratic Synthesizer

#Bioinformatics #Immunoinformatics #AI #Proteomics #Biologics

[Mohamed Elkerdawy @KerdawyBioAI]

Your daily reminder that for most of our history, the goal of an individual was wholeness, not perfection.

[Mohamed Elkerdawy @KerdawyBioAI]

@thisismadani @NatureBiotech @ProfluentBio What a great contribution. Thanks for all humans and institutions involved.

[Mohamed Elkerdawy @KerdawyBioAI]

My 25th year was full of new wonders. I am especially grateful that I could take a closer look at the world views of Alan Watts, Seneca, Marcus Aurelius, Socrates, Hegel, and Carl Jung. No order is meant.

[Calum Bird @calumbirdo]

An optimal language model with 30B parameters can achieve better results than Davinci 175B, while being ~5x cheaper to host and ~2x cheaper to train [1]. I built a calculator so that you can avoid the mistake that cost OpenAI millions👇

[Mohamed Elkerdawy @KerdawyBioAI]

Blessed to have the opportunity to co-author this amazing progress in protein language models.

[Ahmed Elnaggar @Elnaggar_AI]

Introducing Ankh ☥ :
An optimized protein language model (PLM) achieving SoTA performance with < 10% of the top/closest-performing models’ parameters. It is benchmarked on various downstream protein tasks, and tested for protein generation.
https://t.co/7VYwz748TV
(1/n)

[Ahmed Elnaggar @Elnaggar_AI]

In self-supervised learning, increasing the number of parameters to hundreds of billions shows increasing in downstream tasks performance. However, Google revealed that Chinchilla (70B) Outperforms GPT-3 (175B) and Gopher (280B).

Can we do the same for Protein language models?

[Mohamed Elkerdawy @KerdawyBioAI]

If I can promise one thing in this job is that you will work with an interdisciplinary team that will change your perception of science and technology forever. Some relevant examples
@Elnaggar_AI @abs_WaleedM @palepurple__ @Muhamma81966391 @meljendy @azizelgammal

[Mohamed Elkerdawy @KerdawyBioAI]

We are #hiring for a #structural_bioinformatician at @Proteinea. We are localizing the BioTech industry in the MENA region. If you want to join us, we welcome anyone who believes in curiosity, ownership, hustling and being a team player. More info and job post link in the thread.

[Mohamed Elkerdawy @KerdawyBioAI]

@proteinea We are expanding our computational biology team again. This time we are starting with hiring structural bioinformaticians researchers (junior and middle level) (stay tuned for the rest). Link: https://t.co/p3EbWJopDF

[Mohammed AlQuraishi @MoAlQuraishi]

Well deserved!

[Sergey Ovchinnikov @sokrypton]

"scientists solve the protein folding problem" 😀

[Justas Dauparas @JustasDauparas]

Fine-tuning AlphaFold to classify peptide–MHC binding!

[Proteinea @proteinea]

This is as exciting as it sounds: we landed in sunny Saudi Arabia !

[Antonio Regalado @antonioregalado]

Update on CRISPR patent battle.
(Yes, it continues).
Allegation: Zhang/Broad Institute research group learned key detail from Doudna/Berkeley, but didn't admit it. Countercharge: basically, incompetence by Berkeley adapting CRISPR to human cells.
https://t.co/aZfwMJnIyi

[وزارة الاتصالات وتقنية المعلومات @McitGovSa]

مبروك لفريق Pasqal تحقيقهم المركز الثاني في مسار التكنولوجيا من أجل الإنسانية ضمن مسابقة الشركات الناشئة في #ليب22

The second place for Tech for Humanity Award at #LEAP22 goes to Pasqal team, Congratulation

[Eigil Fjeldgren Rischel - abstr/acc @Ayegill]

Making fun of ML for being "just linear algebra" is pretty rich coming from a field that's been publishing fancy variations on linear algebra for the past 212 years.

[Mohammed AlQuraishi @MoAlQuraishi]

This is great news! Our plans for OpenFold won’t change, as having a trainable platform is still incredibly valuable for modifying and building on AF2. The first step ofc is reproducing the AF2 weights independently which is what we’re currently working on.

[Mohamed Elkerdawy @KerdawyBioAI]

@MoAlQuraishi Also please let us know if we can help in terms of computational resources or workforce in the current effort generating the independent weights. We aspire to be part of this great open source effort.

[Mohamed Elkerdawy @KerdawyBioAI]

@MoAlQuraishi At Proteinea, We are finalizing our work on customized antigen-antibody complexes model based on the weights of Rosettafold2. Do you think we can collaborate to replicate this transfer learning methodology with OpenFold, once the weights are available.

[Mohamed Elkerdawy @KerdawyBioAI]

@sokrypton @BrianWeitzner Wow. What a field to be part of. 🎉❤️