RNA-GPS predicts high-resolution RNA subcellular localization and highlights the role of splicing

Kevin Wu, Kevin R. Parker, Furqan M. Fazal, Howard Y. Chang, James ZouRNA 2020 · DOI

Abstract

Subcellular localization is essential to RNA biogenesis, processing, and function across the gene expression life cycle. However, the specific nucleotide sequence motifs that direct RNA localization are incompletely understood. Fortunately, new sequencing technologies have provided transcriptome-wide atlases of RNA localization, creating an opportunity to leverage computational modeling. Here we present RNA-GPS, a new machine learning model that uses nucleotide-level features to predict RNA localization across eight different subcellular locations—the first to provide such a wide range of predictions. RNA-GPS's design enables high-throughput sequence ablation and feature importance analyses to probe the sequence motifs that drive localization prediction. We find localization-informative motifs to be concentrated on 3'-UTRs and scattered along the coding sequence, and motifs related to splicing to be important drivers of predicted localization, even for cytotopic distinctions for membraneless bodies within the nucleus or for organelles within the cytoplasm. Overall, our results suggest transcript splicing is one of many elements influencing RNA subcellular localization.

Introduction: The Cellular Postal System

A cell must precisely localize RNA transcripts to specific compartments—such as the mitochondria or the endoplasmic reticulum—to function correctly. Disruptions in this "postal system" are linked to severe pathologies, including neurodegenerative diseases and cancer. While RNA sequence motifs are known to act as "zip codes," existing computational models have largely been limited to binary (nucleus vs. cytoplasm) predictions and often lack interpretability.

Overview of the eight subcellular compartments and the corresponding transcript counts used for training.

The RNA-GPS Framework

RNA-GPS (Global Positioning System for RNA) addresses these limitations using a Random Forest classifier built on a "segment-aware" featurization strategy. Unlike traditional deep learning models that process RNA as a single continuous string, RNA-GPS divides transcripts into three biologically distinct segments: the 5’ UTR, the Coding Sequence (CDS), and the 3’ UTR.

For each segment, the model calculates the frequency of k-mers (lengths 3, 4, and 5). These counts are normalized by segment length, making the model "length-agnostic" and robust to the high variability in transcript sizes—a common failure point for traditional deep learning models like CNNs or RNNs. This process generates 4,032 features that capture both motif identity and relative position.

The RNA-GPS workflow: segment-wise k-mer counting followed by feature concatenation and Random Forest prediction.

Key Insights: Splicing as a Localization Signal

A primary discovery of this work is the significant role of splicing-related motifs in driving RNA localization. Sequence ablation studies revealed that the 3’ UTR and CDS are the primary "address labels," while the 5’ UTR has a negligible impact. Surprisingly, motifs recognized by splicing factors (e.g., HuR, PTBP1, TIA1) emerged as major drivers of localization across diverse compartments. This suggests that the nuclear splicing machinery plays a dual role in both transcript processing and spatial targeting. Furthermore, the model correctly identified intron retention as a strong signal for nuclear anchoring, mirroring experimental observations.

Results and Performance

RNA-GPS was validated using high-resolution APEX-seq data from HEK293T cells, covering 3,660 transcripts across eight compartments. The model achieved an overall AUROC of 0.77 and an AUPRC of 0.49. In binary tasks (nucleus vs. cytoplasm), it achieved an AUROC of 0.85, significantly outperforming the previous standard, deepLncRNA (0.74). Cross-cell line validation using ENCODE data (HeLa-S3 and K562) further demonstrated the model's superior generalization capabilities.

Conclusion

RNA-GPS demonstrates that a relatively simple, segment-aware machine learning approach can outperform complex deep learning architectures in predicting RNA localization. By highlighting the importance of splicing motifs, this work provides a new lens through which to view the regulation of RNA spatial distribution and its implications for cellular function. Future work may expand this framework to include RNA secondary structure and chemical modifications to further refine localization predictions.

Summary updated

Abstract

Subcellular localization is essential to RNA biogenesis, processing, and function across the gene expression life cycle. However, the specific nucleotide sequence motifs that direct RNA localization are incompletely understood. Fortunately, new sequencing technologies have provided transcriptome-wide atlases of RNA localization, creating an opportunity to leverage computational modeling. Here we present RNA-GPS, a new machine learning model that uses nucleotide-level features to predict RNA localization across eight different subcellular locations-the first to provide such a wide range of predictions. RNA-GPS's design enables high-throughput sequence ablation and feature importance analyses to probe the sequence motifs that drive localization prediction. We find localization informative motifs to be concentrated on 3'-UTRs and scattered along the coding sequence, and motifs related to splicing to be important drivers of predicted localization, even for cytotopic distinctions for membraneless bodies within the nucleus or for organelles within the cytoplasm. Overall, our results suggest transcript splicing is one of many elements influencing RNA subcellular localization.
BibTeX
@inproceedings{wu2020rna,
  author = {Kevin Wu and Kevin R. Parker and Furqan M. Fazal and Howard Y. Chang and James Zou},
  title = {{RNA-GPS predicts high-resolution RNA subcellular localization and highlights the role of splicing}},
  booktitle = {RNA},
  year = {2020},
  doi = {10.1261/rna.074161.119},
}
Figure