Hussein Mahdi

classes, but message is clear → chemistry drives metabolism.

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.

🏅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

classes, but message is clear → chemistry drives metabolism.

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.

🏅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

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.

🚀 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

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

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.

🚀 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

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

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

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?

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

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.

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

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.

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

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.

Half of all patients take multiple medications, but most don't know about dangerous drug interactions.
The stats are scary:

106K deaths/year from drug reactions
20% caused by drug interactions
Costs up to $180B annually

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

recyclable and maintainable, less dependent on dependencies, and resistant to obsolescence 😑😑

In fact, some people even overuse external libraries and services for a project with less than a thousand users, and their justification for that is that clean code must be strong and coherent, and they forget that it must be easy and simple to understand,

High-level overview of the architectural design principles guiding the new game engine.

#gamedev #badcatgame #catgame #softwarearchitecture

If you value a high quality, regulated food supply, contact your reps!

Science Feed Rules: 🧪 No misinformation 🧪 Keep posts relevant to science 🧪 Use alt text 🧪 Credit images 🧪 Give context to links to external sites 🧪 Have fun!

Welcome to the Science feed!

Please read our FAQs for instructions for how to be added as a contributor: bossett.io/science-feed/

Mod introductory posts linked below⬇️

Use the test tube emoji on posts you want to appear in the feed🧪

Please like the feed and make sure you follow our feed rules:

Using cutting-edge "sparse hypergraph" technology, it models complex drug relationships while filtering out noise to focus on what matters most for patient safety.

This means: fewer preventable hospital visits, safer prescriptions, faster drug development!

#AI #Healthcare #DrugSafety #ML

Oxford's 2022 SPARSE neural network can predict dangerous drug interactions before they happen - potentially saving lives by identifying risky medication combinations early.

My analysis paper
github.com/Hu8MA/Papers...

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

Reviewed HGDrug—hypergraph learning for drug interactions!

Cool:
- BRICS fragments + biological nets → multi-task predictions (DDI/DTI/DSI).
- RDF-style reasoning: "Drug A → Fragment X → Side Effect Y".
- 97.6% AUROC

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

Issue:
•Complexity ≠ performance: Heavy architecture, tiny gains.
•Chemical fragments rule: Biological nets add little predictive power.

Paper: journals.plos.org/ploscompbiol
review:
github.com/Hu8MA/Papers...

#MedChem #OpenScience #AIforGood #Cheminformatics

Reviewed HGDrug—hypergraph learning for drug interactions!

Cool:
- BRICS fragments + biological nets → multi-task predictions (DDI/DTI/DSI).
- RDF-style reasoning: "Drug A → Fragment X → Side Effect Y".
- 97.6% AUROC

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