@bioalgorithmist

Scientist, ML at Tessera | Algorithms for Gene Writing™ | PhD from Salis Lab at Penn State | threads/X/bsky @bioalgorithmist | Opinions are Mine Only

 Boston, MA
github.com/ayaanhossain/
Joined April 2017
  • Tweets552
  • Following263
  • Followers284
Very happy to release our latest paper from @hsalis Lab in collaboration with @klavins Lab at UW on "Automated design of thousands of nonrepetitive parts for engineering stable genetic systems", now published in Nature Biotechnology! 1/18 nature.com/articles/s41587-0…
4
7
2
33
The LLM Paradox: These things can generate photorealistic images, compose symphonies, and produce cinematic videos, yet somehow struggle to write a paragraph of genuinely good prose. You'd think a Large Language Model would at least be good at the language part.
71
Ayaan Hossain retweeted
BURKOV
@burkovFeb 12
Human coders know they lost, but they keep fighting with windmills because they aren't ready to accept it.
15
8
1
88
15,752
Our work developing a parts list of promoters and gRNA scaffolds for mammalian genome engineering and molecular recording is now out @NatureBiotech
Nature Biotechnology
@NatureBiotech11 Nov 2025
A parts list of promoters and gRNA scaffolds for mammalian genome engineering and molecular recording go.nature.com/49eTPCu
2
10
45
14,189
A parts list of promoters and gRNA scaffolds for mammalian genome engineering and molecular recording go.nature.com/49eTPCu
1
22
2
90
23,372
🧠 Why do smart scientists feel stupid when reading papers? Because nobody teaches you HOW to read them efficiently. This 3-pass system will change how you approach every paper: 🧵
4
30
3
139
14,421
Ayaan Hossain retweeted
Nic Fishman@njwfish26 May 2025
🚨 New preprint 🚨 We introduce Generative Distribution Embeddings (GDEs) — a framework for learning representations of distributions, not just datapoints. GDEs enable multiscale modeling and come with elegant statistical theory and some miraculous geometric results! 🧵
6
133
6
745
81,151
New lab preprint! 🚀 Modeling complex data distributions is tough. We designed GDEs, a new framework that tackles this head-on! GDEs generalize across text, images & MANY bio apps (think virtual cells, spatial bio, viral genome tracking). Thread 👇
3
6
1
60
18,927
If you have a solid strategy and a small amount of compute, you can go pretty far. If you have huge clusters of GPUs and no strategy, your only achievement will be burning capital.
33
67
10
812
42,399
Ayaan Hossain retweeted
Asimov
@AsimovBio13 Nov 2024
This is a great paper from the @hsalis lab. - Measure the decay rates of 50,000 mRNAs in bacteria. - Use biophysical models + ML to build models of mRNA stability. - Profit. And a good reminder of what's possible when one turns a biological problem into a sequencing problem!
1
24
87
7,995
I am pleased to announce our latest publication ‘Predicting synthetic mRNA stability using massively parallel kinetic measurements, biophysical modeling, and machine learning’ in @NatureComms
1
2
4
165
Designing synthetic mRNAs 🧬 for optimal stability and expression? Checkout our work led by @DanielCetnar on biophysics infused machine learning approaches for delineation of mRNA degradation kinetics ⚡ out now in @NatureComms nature.com/articles/s41467-0…

Predicting synthetic mRNA stability using massively parallel kinetic measurements, biophysical...

Nature Communications - The authors carried out massively parallel mRNA decay measurements on over 50,000 mRNAs, revealing the interactions controlling mRNA stability in bacteria. Biophysical...

nature.com
1
8
33
1,943
We applied rational learn-by-design 🔢 methods to create a maximally informative library 📚, coupled that with high throughput, barcoded, massively parallel reporter assays to decrypt major design rules using Gradient Boosted Trees 🌳!
1
1
119
Congratulations to co-authors @grace_vezeau and massive thanks to Prof. @hsalis for making us part of this exciting project! 🙏
1
93
Ayaan Hossain retweeted
Akos Nyerges@AkosNyerges15 Jul 2024
Thanks for the highlight @doescience! Read more about our research: nature.com/articles/s41586-0… biorxiv.org/content/10.1101/…

A swapped genetic code prevents viral infections and gene transfer

Nature - A study details the creation of an Escherichia coli genetically recoded organism that is resistant to viral infection, and describes a further modification that keeps the organism and its...

nature.com
DOE Office of Science@doescience14 Jul 2024
Engineered bacteria are great for making biofuels and other bioproducts, but you don’t want those changes escaping to the wild. Scientists @Harvard & @WUSTLmed engineered the E. coli genome to keep useful edits from spreading in an uncontrolled way: energy.gov/science/ber/artic…
A bacterial virus inserting genetic material inside a bacterial cell during infection. Image courtesy of Behnoush Hajian.

ALT A bacterial virus inserting genetic material inside a bacterial cell during infection. Image courtesy of Behnoush Hajian.

6
29
4,913
Supremely entertaining paper, "ChatGPT is bullshit": link.springer.com/article/10…
1
3
79
Ayaan Hossain retweeted
Akos Nyerges@AkosNyerges17 Jun 2024
In our newest preprint, • we explore the effects of synonymous genome recoding, and • construct & troubleshoot a synthetic 57-codon E. coli genome using multi-omics, editing, and laboratory evolution. 1/n
5
48
1
147
26,042
Ayaan Hossain retweeted
Akos Nyerges@AkosNyerges17 Jun 2024
After almost two years of optimization, SynOMICS's plasmid system is a fusion of @dbikard @MarraffiniLab's Cas9 & the pORTMAGE plasmids & ELSA/nonrepetitive parts by @hsalis @bioalgorithmist Alex Reis & co I highly recommend using multiplexed sgRNAs from nature.com/articles/s41587-0…
1
2
8
957
Please RT: If there's an old unmaintained software that is prohibitively slow and is only one of its kind do you not benchmark your new tool against it? How to properly address reviewers and frame this issue in paper?
2
2
3
1,716
Ayaan Hossain retweeted
Alex Reis@alexcampreis28 Feb 2024
If you are using Language Models to predict gene expression, you better compare with latest state-of-the-art biophysical or ML models
Howard Salis@hsalis28 Feb 2024
Replying to @anshulkundaje
Our bacterial Promoter Calculator model has 348 parameters and predicts cont-val TX rates with R^2=0.80 (other factors constant). Our bacterial RBS Calculator has 12 free parameters and predicts cont-val TL rates with R^2=0.75. The bar is pretty high, but they don’t compare.
1
6
30
11,532
Ayaan Hossain retweeted
David Ding@ding_david23 Feb 2024
Do you need a billion parameters and millions of sequences for #proteindesign? Maybe not! See our paper: shorturl.at/novMO We show that interpretable, non-autoregressive structure-based protein design can work! Original thread: shorturl.at/cdgCP
3
31
2
187
23,051
Load more