Mehdi Yazdani

Graduate Research Assistant @UCF

[Mehdi Yazdani @yazdanimehdi]

I will be at @neurips in San Diego this year, and I am excited to connect with old friends and meet new ones. If you are attending and want to connect, please reach out. Also, I am actively looking for postdoc opportunities in ML for scientific discovery and drug discovery!

[Mehdi Yazdani @yazdanimehdi]

@Micro_Yunha Hi Dr Hwang, I recently defended my phd in CS working on ML for drug discovery and biological sequence modeling. I will also be at NeurIPS Dec 5 to 7 and would love to chat about postdoc opportunities in your group if you have time. I can also send an email with more details.

[Artem Moskalev @artemmoskalev]

Symmetries+biology is a beautiful combination!

[Biology+AI Daily @BiologyAIDaily]

Equi-MRNA: Protein Translation Equivariant Encoding for mRNA Language Models

1. Computational biologists have introduced Equi-mRNA, a new language model for mRNA that uses group-theoretic priors to explicitly encode the inherent symmetries of synonymous codons. This innovative approach allows the model to learn biologically grounded representations.

2. On downstream property-prediction tasks, Equi-mRNA shows significant improvements, with up to a 10% gain in accuracy over traditional baseline models on tasks including expression, stability, and riboswitch switching.

3. In sequence generation, the model produces highly realistic mRNA constructs, achieving a roughly 4x improvement in fidelity and better preserving functional properties compared to vanilla models.

4. The model is also highly efficient, with a 15M parameter variant outperforming larger architectures while using only about 30% of the parameters of a comparable baseline.

5. Interpretability analyses reveal that the model's learned codon-rotation patterns align with known biological phenomena like GC-content biases and tRNA abundance, providing new insights into codon usage.

📜Paper: https://t.co/9KK1gVWrxt
#computationalbiology #mRNA #AIinScience #Bioinformatics #LanguageModels

[Towards Data Science @TDataScience]

Regularization might look like the easiest way to balance accuracy and fairness — but is it the most principled? Mehdi Yazdani explores how bilevel optimization offers a cleaner, more modular alternative that can actually improve convergence and maintain Pareto-optimal trade-offs.

https://t.co/AF7kLFUQZI

[Mehdi Yazdani @yazdanimehdi]

I just published Bilevel Optimization: A Better Way to Balance Competing Objectives in Neural Nets https://t.co/SZciYzNwxW

[Artem Moskalev @artemmoskalev]

ICLR update 🚨 What's missing in existing deep learning methods for efficient modeling of RNA molecules? We spent a year investigating, and I’m excited to share that our work led to two papers accepted at ICLR:

1. HELM: Hierarchical Encoding for mRNA Language Modeling https://t.co/b3VlrPUS7R

2. Beyond Sequence: Impact of Geometric Context for RNA Property Prediction https://t.co/49wlfKO5Ib

[Artem Moskalev @artemmoskalev]

Don't treat the language of biology as natural language! Biology speaks in hierarchical patterns that natural language models don't fully capture. Meet HELM: a novel approach to train LMs that aligns with the intrinsic hierarchy of mRNA sequences.

https://t.co/CqKPxRdQFy

1/5

[Mehdi Yazdani @yazdanimehdi]

I am speaking at Research in Computational Molecular Biology 2024. Please check out my talk if you're attending the event! @RECOMBconf #RECOMB2024 #RECOMB - via #Whova event app

[Mehdi Yazdani @yazdanimehdi]

Introducing DeepDrugDomain: A cutting-edge toolkit for Drug-Target Interaction & Affinity Prediction. Streamlined preprocessing, advanced modeling capabilities, and more https://t.co/sOuhzt2aae) #Bioinformatics #MachineLearning #DrugDiscovery

[bioRxiv Bioinfo @biorxiv_bioinfo]

BindingSiteAugmentedDTA: Enabling A Next-Generation Pipeline for Interpretable Prediction Models in Drug-Repurposing https://t.co/t3Q0ZgXCtr #biorxiv_bioinfo

[bioRxiv @biorxivpreprint]

BindingSiteAugmentedDTA: Enabling A Next-Generation Pipeline for Interpretable Prediction Models in Drug-Repurposing https://t.co/tzTnT5Onvf #bioRxiv

[doctick @doctick]

#NLP "AttentionSiteDTI: an interpretable graph-based model for drug-target interaction prediction using NLP sentence-level relation classification" https://t.co/G61p8ikWFG

[Mehdi Yazdani @yazdanimehdi]

Our AOTA method for drug-target interaction "AttentionSiteDTI: an interpretable graph-based model for drug-target interaction prediction using NLP sentence-level relation classification." has been published in Briefings in Bioinformatics.
https://t.co/XSDXV6Ue67

[bioRxiv Bioinfo @biorxiv_bioinfo]

AttentionSiteDTI: Attention Based Model for Predicting Drug-Target Interaction Using 3D Structure of Protein Binding Sites https://t.co/YA7lCJmQlJ #biorxiv_bioinfo

[Mehdi Yazdani @yazdanimehdi]

AttentionSiteDTI: Attention Based Model for Predicting Drug-Target Interaction Using 3D Structure of Protein Binding Sites https://t.co/53fIUG0sGO