Paper on spatial transcriptomics imputation published in Briefings in Bioinformatics!

Thrilled to announce my latest publication in Briefings in Bioinformatics!

Our study, Spatial information matters: are traditional imputation methods effective for spatial transcriptomics data?, explores the critical role of spatial context in improving imputation for high-resolution spatially resolved transcriptomics (SRT).

🔍 Why this matters

SRT technologies provide near single-cell resolution, unlocking unprecedented biological insights. Yet, these datasets are often sparse and plagued by dropout events, making accurate interpretation challenging. Traditional imputation methods fall short when applied to emerging SRT platforms.

💡 What we built

We benchmarked seven state-of-the-art imputation methods across five SRT platforms and 23 datasets. Our findings revealed that no single method consistently excels—highlighting the need for spatially aware approaches.

To address this, we developed SpaMean-Impute, a novel imputation method that integrates spatial information to better recover missing values and detect valid dropouts.

🧪 Key Results

• 📈 Performance gains over SOTA methods:
  • +16.15% ARI
  • +18.45% NMI
  • +18.96% AMI
  • +13.98% HOMO
• ⚡ Efficiency boost: ~33× faster and ~1500 MB less memory usage compared to deep learning-based imputation methods.

🌐 Impact

SpaMean-Impute demonstrates how leveraging spatial context can significantly improve both accuracy and efficiency in SRT data analysis—paving the way for more reliable biological discoveries.

Grateful to my co-authors Riasat Azim and Swakkhar Shatabda for their collaboration and insights. Excited to see how this work contributes to advancing computational biology!

Check out the full paper and source code here:
Paper Link: https://doi.org/10.1093/bib/bbag027

Source Code: https://github.com/FahimHafiz/SpaMean-Impute