@MIT trained Neuroscientist interested #Neural_Nets, #BioAI, #NeuroAI 🧠⚡🤖 substack.com/@bioai2neuro 🎯 aibio28ai@gmail.com
Texas, United States
- Tweets 3,154
- Following 7,957
- Followers 15,778
- Likes 7,483
Interesting! This sounds like a paradox !
Higher-resolution images can make an AI model less accurate.
In a recent Nature Communications study, Ferreira and colleagues imaged the same cell nuclei at different magnifications and trained segmentation models at each resolution.
The models performed better at 20× than at 60×. Acquiring the same field of view also took 9.5 seconds instead of 36.5.
The explanation is useful well beyond microscopy.
A network uses a limited region of the image to make each local prediction. At higher magnification, a nucleus occupies more pixels and can extend beyond that region. The model loses the context needed to identify its boundaries.
Reducing the image scale can bring that context back into view.
For me, the practical lesson is that measurement scale belongs in model design. Before investing in more detailed data, check whether that detail actually helps the scientific task.
I explore this and other key ideas in the latest Discovery at Scale, looking at the choices about data, targets and context that can improve scientific ML before we reach for a larger model.
Read the full issue: bravoabad.substack.com/p/wha…
1
1
4
789
@BioAI_Neuro retweeted
Please retweet!! Thank you
Are you a journal editor, editorial board member, or involved in the editorial process and looking to commission a timely review or perspective on #BioAI or #NeuroAI?
I’d be happy to explore opportunities to write a review covering emerging intersections of AI, neuroscience, biology, and biomedical research.
If you are interested in inviting a review, perspective, or commentary, feel free to reach out or DM me. I’d be glad to discuss potential topics and scope.
Here is a sample article that received lot of attention here on @X
nitter.net/compose/articles/edit/…
Please retweet!! Thank you
#BioAI #NeuroAI #Neuroscience #Neurotechnology #BrainAI #ScientificPublishing #AcademicPublishing
Curious about synthetic data in BioAI and NeuroAI?
Simulators, virtual cells, model collapse, and one model that went from R² 0.999 to 0.15 on real neurons.
Here it is 👇
nitter.net/BioAI_NeuralNet/status…
#BioAI #NeuroAI #Foundationmodels
6
8
1,088
Heart Transplants And Crisscross Aging
Hearts from young donors age rapidly in the bodies of older recipients, and older hearts are rejuvenated inside young people
go.nature.com/4rzzsHf
1
3
705
Ability of a Structural World Model to Detect Cryptic Pockets from Apo Structure
biorxiv.org/content/10.64898…
Summary:
This study presents a method for identifying cryptic drug-binding pockets directly from a single apo protein structure, without requiring molecular dynamics simulations, conformational sampling, co-folding predictions, or external pocket-detection tools. The approach uses a proprietary structural world model that generates a per-residue latent representation from protein coordinates. This latent state is read out as a cryptic-lining score, suggesting the model implicitly captures conformational flexibility that other methods must explicitly simulate.
Predicted pockets are constructed as residue sets rather than fixed geometric spheres. High-scoring residues seed candidate pockets, which are expanded into distinct, non-overlapping predictions through a residue-growth strategy and geometric non-maximum suppression.
On CryptoBench (231 test proteins), the model achieves 84.8% top-1 and 95.2% top-5 localization accuracy. Similar performance is observed on a CryptoBank subset, with 84.6% top-1 and 99.0% top-5 accuracy. Residue-level classification is strong (AUC = 0.8465), although exact pocket-boundary recovery remains more challenging under stricter overlap criteria.
The method generalizes well to unseen proteins, recovering the known WRN helicase allosteric site at rank 1 across multiple apo structures after all WRN proteins were removed from training. A key advantage is its ability to place the true cryptic site at rank 1, addressing a common limitation of ensemble-based approaches.
Compared with single-structure baselines such as P2Rank, DeepPocket, PocketMiner, and fpocket, the model achieves substantially higher top-1 hit rates. It also complements the ensemble-based method OpenDDE, identifying many cryptic sites missed by OpenDDE while retaining all of OpenDDE’s top-5 hits. Overall, the work demonstrates that latent representations learned by a structural world model can effectively detect and rank cryptic pockets from apo structures alone, with performance improving as training data increases.
#DrugDiscovery #CrypticPockets #BioAI #AIforScience #ProteinStructure
2
3
6
654
Molecular heterogeneity of Aging across populations.
Aging does not appear to follow the same molecular script for everyone, according to an 8-year study of more than 300 women.
The findings in Science reveal that individual molecular trajectories of aging can diverge substantially from population-wide patterns and are shaped not only by genetics but also by factors such as circadian rhythm, seasonality, and environmental exposures. Learn more: scim.ag/4ikMUfG
1
2
9
1,339
@BioAI_Neuro retweeted
Curious about synthetic data in BioAI and NeuroAI?
Simulators, virtual cells, model collapse, and one model that went from R² 0.999 to 0.15 on real neurons.
Here it is 👇
nitter.net/BioAI_NeuralNet/status…
#BioAI #NeuroAI #Foundationmodels
2
10
18
2,961
Single-Cell Mapping of Transcriptomic Vulnerability in Brain Disorders
Nature research paper: Single-cell atlas of transcriptomic vulnerability across brain disorders
go.nature.com/4hrjOJg
1
2
8
1,235
Thank you! Please shsre your feedback . If you want me to write something that you enjoy reading, please let me know.
1
1
15
Completely agree! There is no replacement for experimental data. Synthetic data is just an option when real biological data id hard to collect due to technical challenges.
1
22
@BioAI_Neuro retweeted
🧠Great post on Neuronal modeling, with refs and code
"Synthetic data will be part of every serious BioAI and NeuroAI foundation model, because real data cannot cover the space these models are asked to predict. The open question is not whether to use it but how to keep it honest.
1
1
6
754
A multimodal, all-optical platform for linking cell identity to metabolic function in intact tissues
nature.com/articles/s41592-0…
2
6
1,437