LPI-KTASLP: Prediction of LncRNA-Protein Interaction by Semi-Supervised Link Learning With Multivariate Information

Cong Shen, Yijie Ding, Jijun Tang, Limin Jiang, Fei Guo

Abstract

Long non-coding RNAs (lncRNAs) are essential regulators of gene expression, primarily functioning through interactions with specific proteins. Mapping these interactions experimentally is costly and time-consuming. This paper presents LPI-KTASLP, a novel computational framework that predicts lncRNA-protein interactions (LPIs) by treating the task as a link-prediction problem in a bipartite network. The framework integrates multivariate biological data—including Gaussian interaction profiles, sequence features, and expression data—using Kernel Target Alignment (KTA) to weight data sources optimally. By employing semi-supervised link prediction (SLP) with low-rank approximations, the model efficiently utilizes both known and unknown interactions. Experimental results on benchmark datasets demonstrate that LPI-KTASLP significantly outperforms existing state-of-the-art methods, achieving an AUPR of 0.6148 on the NPInter V2.0 dataset.

Introduction

Long non-coding RNAs (lncRNAs) were historically overlooked as "junk DNA," but they are now recognized as critical regulators of biological processes and disease progression. Their regulatory roles are largely mediated by binding to proteins. However, the vast majority of potential lncRNA-protein interactions remain unmapped due to the limitations of laboratory techniques.

Computational "in silico" methods offer a scalable alternative, yet they face two major hurdles: effectively combining heterogeneous data types (Data Integration) and handling the lack of confirmed non-interactions in biological datasets (The "Unlabeled" Problem).

Methods: The LPI-KTASLP Framework

LPI-KTASLP addresses these challenges through a multi-step computational pipeline focused on multivariate kernel fusion and efficient semi-supervised learning.

Multivariate Kernel Fusion and KTA

The model constructs multiple similarity matrices, or "kernels," for both lncRNAs and proteins. These include Gaussian interaction profiles (GIP), sequence features, and expression data. To ensure the most relevant data has the greatest influence, the authors use Kernel Target Alignment (KTA). KTA automatically weights each kernel based on how well it aligns with known interaction patterns, ensuring that the most reliable information carries more weight in the final prediction.

Multivariate kernel fusion and Kernel Target Alignment (KTA) for protein space.

To handle the "unlabeled" problem, the framework utilizes semi-supervised learning. Unlike supervised methods that require explicit negative samples, SLP propagates interaction scores across the entire bipartite network, leveraging the global structure to identify likely new links. To manage the computational complexity of large networks, the authors implement low-rank approximations and eigendecomposition, making the model significantly faster than traditional methods.

Comparison between supervised and semi-supervised learning paradigms for LPI prediction.

Experimental Results

LPI-KTASLP was rigorously evaluated on two primary datasets: the NPInter V2.0 benchmark (4,158 interactions) and the larger Zheng dataset (4,467 interactions).

  • Superior Accuracy: On NPInter V2.0, LPI-KTASLP achieved an AUPR of 0.6148, a substantial improvement over previous methods like LPLNP (0.4584) and RWR (0.2827).
  • Metric Selection: The authors emphasize the Area Under the Precision-Recall curve (AUPR) as a more critical metric than the standard AUC for LPI prediction. Since true interactions are rare, AUPR more accurately reflects the model's ability to minimize false positives.
  • Literature Validation: Case studies on specific proteins showed that 60-70% of the top 20 predicted novel interactions were already supported by independent biological literature, confirming the model's real-world utility.

Conclusion and Future Work

LPI-KTASLP provides a robust and efficient tool for mapping the regulatory landscape of the cell. While it sets a new performance standard, the authors identify several areas for future improvement, including the incorporation of 3D folding information for RNAs and proteins, expanding the variety of multivariate kernels available for fusion, and exploring non-linear strategies for fusing heterogeneous data sources.

Summary updated

Abstract

Long non-coding RNA, also known as lncRNA, is a series of single-stranded polynucleotides (no less than 200 nucleotides each), consisting of non-protein coding transcripts. LncRNA plays a crucial role in regulating gene expression, during the transcriptional, post-transcriptional, and epigenetic processes. This is achieved by lncRNA interacts with the corresponding RNA-binding proteins. It has been drawn to a lot of attention that the reduction of the excessive laboratory cost and the increase in speed and accuracy gains benefits from the employment of computational intelligence in lncRNA-protein interaction (LPI) identification. Although numerous pertinent in silico studies of LPI prediction have been proposed, there is still room for enhancing the accuracy of the existing LPI prediction methods. In this paper, we have proposed a novel method for identifying LPI with kernel target alignment based on semi-supervised link prediction (LPI-KTASLP), which adopts multivariate information to predict lncRNAs-proteins interactions. To integrate the heterogeneous kernels, kernel target alignment has been applied to deal with kernel fusion. We have calculated the low-rank approximation matrices of lncRNA and protein, where eigendecomposition is used to reduce computing pressure. The prediction model has been obtained by producing the ultimate LPI prediction matrix. Experimental results show that the prediction ability of the LPI-KTASLP algorithm has surpassed many other LPI prediction schemes. Our method of lncRNA-protein interaction prediction has been evaluated on a standard benchmark dataset of LPIs. We have observed that the highest AUPR of 0.6148 is obtained by our proposed model (LPI-KTASLP). This is superior to the integrated LPLNP (AUPR: 0.4584), the RWR (AUPR: 0.2827), the CF (AUPR: 0.2357), the LPIHN (AUPR: 0.2299), and the LPBNI (AUPR: 0.3302). It is very encouraging that most of the LPI predictions have been confirmed to be close to relevant concentrations.
BibTeX
@article{shen2019lpi,
  author = {Cong Shen and Yijie Ding and Jijun Tang and Limin Jiang and Fei Guo},
  title = {{LPI-KTASLP: Prediction of LncRNA-Protein Interaction by Semi-Supervised Link Learning With Multivariate Information}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2894225},
}
Figure