Lightnews — Scholar-powered news

Hussein Mahdi @hussein16mahdi.bsky.social · 1d

Fathers of languages

#Science #SciComm #Software#dev #AcademicSky

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 4d

And this isn’t a one-shot: results come from 3 model configs × 2 seed partitions → stable + robust.
Even vs. papers like DeepDDI, NMDADNN, SumGNN, HetDDI (with heavy KGs + multi-modal data), this SMILES-only design holds up.

1 1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 4d

Update: my second DDI network (multi-class, 86 types) is showing strong preliminary results:
• ~85% Top-1
• ~98% Top-3
• Just 18 mins training on free Colab ⚡

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 4d

Next: improve rare classes, but message is clear → chemistry drives metabolism

Hussein Mahdi @hussein16mahdi.bsky.social · 4d

And this isn’t a one-shot: results come from 3 model configs × 2 seed partitions → stable + robust.
Even vs. papers like DeepDDI, NMDADNN, SumGNN, HetDDI (with heavy KGs + multi-modal data), this SMILES-only design holds up.

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 4d

Update: my second DDI network (multi-class, 86 types) is showing strong preliminary results:
• ~85% Top-1
• ~98% Top-3
• Just 18 mins training on free Colab ⚡

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1 1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

classes, but message is clear → chemistry drives metabolism.

1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

And this isn’t a one-shot: results come from 3 model configs × 2 seed partitions → stable + robust.
Even vs. papers like DeepDDI, NMDADNN, Next: improve rare
SumGNN, HetDDI (with heavy KGs + multi-modal data), this SMILES-only design holds up.

1 1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

🏅Update: my second DDI network (multi-class, 86 types) is showing strong preliminary results:
• ~85% Top-1
• ~98% Top-3
• Just 25 mins training on free Colab ⚡️

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

classes, but message is clear → chemistry drives metabolism.

1

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

And this isn’t a one-shot: results come from 3 model configs × 2 seed partitions → stable + robust.
Even vs. papers like DeepDDI, NMDADNN, Next: improve rare
SumGNN, HetDDI (with heavy KGs + multi-modal data), this SMILES-only design holds up.

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 11d

🏅Update: my second DDI network (multi-class, 86 types) is showing strong preliminary results:
• ~85% Top-1
• ~98% Top-3
• Just 25 mins training on free Colab ⚡️

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 17d

Beyond strong benchmarks, we extended the model to multi-task learning: not only detecting interactions but also classifying their type ; especially metabolic ones vital for clinicians.

1 1

Reposted by Hussein Mahdi

Hussein Mahdi @hussein16mahdi.bsky.social · 17d

🚀 Exciting milestone at Era Vision Research Center!
We reproduced and advanced a hypergraph-based neural network for drug–drug interaction (DDI) prediction.

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 17d

Bridging AI innovation with clinical decision support for interpretable, clinically relevant impact.

Hussein Mahdi @hussein16mahdi.bsky.social · 17d

Beyond strong benchmarks, we extended the model to multi-task learning: not only detecting interactions but also classifying their type ; especially metabolic ones vital for clinicians.

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 17d

🚀 Exciting milestone at Era Vision Research Center!
We reproduced and advanced a hypergraph-based neural network for drug–drug interaction (DDI) prediction.

#AI #MachineLearning #DeepLearning #Research #AcademicSky #Science #SciComm #Software #dev #AcademicSky #phylogenetics

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · 25d

12 experiments, 1 neural network… my Colab tabs look like NASA mission control 🚀
Each seed is fighting for the title of Best Epoch Champion 🏆😂

#AI #MachineLearning #DeepLearning #AI #ML #PharmaTech #Research #AcademicSky #Science #SciComm #Software #dev

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 18

New drugs face the "cold-start problem" - no historical data means accuracy drops from 95% to 80%. Still working on that one! 🤔

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 18

can predict what happens when 3, 4, or 5+ medications interact simultaneously - because that's reality 💪 for many patients.

Traditional methods could only say "Drug A + Drug B = problem." Now we can model entire medication cocktails.
The challenge?

1

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 18

Here's the wild part about modern drug interaction prediction:

we're moving beyond simple pairwise relationships 🧐👨‍💻

New "hypergraph" approaches 🥇

#AI #ML #PharmaTech #Research #AcademicSky #Science #SciComm #Software #dev

1 1

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 16

The stakes? With thousands of drugs in use, potential interactions number in the millions. Computational prediction isn't just helpful anymore - it's literally life-saving.

1

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 16

These AI systems learn directly from molecular structure, discovering dangerous drug interactions that human experts might miss. Knowledge graphs now integrate BILLIONS of data points from medical databases, connecting drugs, proteins, and diseases in unprecedented ways.

https://www.google.com/url?sa=i&url=https%3A%2F%2Fwww.mdpi.com%2F2076-3417%2F13%2F5%2F2842&psig=AOvVaw2bBEQmWBl6Ymhgavnlj_IC&ust=1758147514703000&source=images&cd=vfe&opi=89978449&ved=0CBgQjhxqFwoTCJDmioWo3o8DFQAAAAAdAAAAABBA

1

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 16

Graph Neural Networks are transforming drug safety prediction by treating molecules like social networks - atoms as people, bonds as relationships.

#AI #ML #PharmaTech #Research #AcademicSky #Science #SciComm #Software #dev

1 3

Hussein Mahdi @hussein16mahdi.bsky.social · Sep 9

As people live longer and take more meds, this problem is exploding. Traditional safety methods aren't keeping up.

Good news? Government agencies are finally investing in AI to predict these deadly combinations before they happen.
Know your meds.
Ask questions. Stay safe.