Accelerated Transfer Learning for Protein-Protein Interaction Prediction

Barnes, Bradley

doi:10.22215/etd/2018-12707

Cited by 3 publications

(7 citation statements)

References 72 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The approach works on the underlying hypothesis that across various proteins, almost similar local subsequences of amino acids would be yielding a 3D structure, and these structures will lead to forming interactions [13]. It works on the principle that the input candidate protein pair A-B are similar sequence to variety of protein pairs that are known to interact (e.g.…”

Section: Principlementioning

confidence: 99%

“…As an example, PIPE uses a PAM120 score threshold of 40 for Homo Sapiens and 35 for Saccharomyces cerevisiae [47]. Iterating through the full length of both protein A and protein B leads to the creation of a 2D matrix that is as wide as the amino acid length of protein A minus nineteen, and as tall as the amino acid length of protein B minus nineteen [13]. Nineteen is subtracted due to the first twenty amino acids in the window being summarized by the first element.…”

Section: Core Algorithmmentioning

confidence: 99%

“…The height of a peak at location (i, j) represents to weight of evidence that subsequences WAi and WBj support a physical interaction between proteins A and B. By aggregating the height of all locations in the landscape, it arrives at a total score for the putative interaction between proteins A-B [13]. It is also possible to estimate the site of interaction within each protein by examining the landscape itself, which is discussed in the coming sections, known as PIPE-Sites.…”

Section: Output Datamentioning

confidence: 99%

“…The second method of summarizing the landscape into a single score is known as the similarity-weighted (SW) method [13]. This method was developed because certain sliding windows are seen in a very large number of proteins but are not responsible for supporting or mediating interactions.…”

Section: Sw-score and Sw-landscapementioning

confidence: 99%

See 3 more Smart Citations

Sequence-based Protein Interaction Site Prediction using Computer Vision and Deep Learning

Purohit¹

View full text Add to dashboard Cite

Computational prediction of protein-protein interaction (PPI) from protein sequence is important as many cellular functions are made possible through PPI. The Protein Interaction Prediction Engine (PIPE) software suite was developed at Carleton University for such predictions. This thesis aims to conduct a thorough performance assessment of the PIPE-Sites predictor through the use of a large high-quality set of known PPI sites. The results show that PIPE-Sites has relatively low accuracy even after retuning the inherent hyperparameters of the method. Furthermore, PIPE-Sites are shown to be ineffective when applied to similarity-weighted score data. Thus, three new sequence-based methods of predicting PPI sites are proposed and evaluated, including the Panorama, BrightSpot, and ClusterNet methods. The new methods leverage similarityweighted score data to further increase performance. Ultimately, ClusterNet significantly outperforms the other methods over two different performance metrics when evaluated on both human and yeast data PPI site data.

show abstract

Section: Principlementioning

confidence: 99%

Section: Core Algorithmmentioning

confidence: 99%

Section: Output Datamentioning

confidence: 99%

Section: Sw-score and Sw-landscapementioning

confidence: 99%

See 2 more Smart Citations

Sequence-based Protein Interaction Site Prediction using Computer Vision and Deep Learning

Purohit¹

View full text Add to dashboard Cite

show abstract

“…The generated score can be used to rank all computed pairs by likelihood of a positive PPI, or for a given threshold a binary decision can be made. (Figure reproduced from[44])…”

mentioning

confidence: 99%

Improving the Protein-Protein Interaction Prediction Engine (PIPE) with Protein Physicochemical Properties

Jary¹

View full text Add to dashboard Cite

Protein-protein interactions (PPI) serve an important role in both protein and cell function. They are difficult and time consuming to determine experimentally and thus benefit from in silico prediction methods. This thesis improves a high throughput, sequence-based protein-protein interaction prediction method called the protein-protein interaction engine (PIPE). The initial contribution was in developing comprehensive documentation on how to run and use PIPE. Next, a Python implementation of the scoring of PIPE was developed. A software framework was created for the systematic exploration of how physicochemical properties can improve PPI prediction. Subsequently, a sequence-based solvent accessibility approach was integrated with PIPE, improving PPI prediction recall by 0.9% from 73.8% to 74.7% at 90% precision. Finally, 166 different sequence-based physicochemical properties were generated using the ProtDCal software tool and were integrated with PIPE using the framework developed in this thesis. The best of these properties improved the recall of PIPE by 2% at 90% precision. This improvement was shown to be statistically significant and was confirmed on a larger test set including 10,000 protein pairs known to interact and 10,000 randomly selected pairs, assumed not to interact.

show abstract

Employing Machine Learning Techniques to Detect Protein-Protein Interaction: A Survey, Experimental, and Comparative Evaluations

Taha

2023

Preprint

View full text Add to dashboard Cite

This survey paper provides an in-depth analysis of various machine learning techniques and algorithms that are utilized in the detection of PPI (Protein-Protein Interactions). For every technique examined, the paper evaluates its efficiency, shortcomings, possibilities for enhancement, and outlook for the future. A major challenge in current survey papers focusing on machine learning algorithms for PPI identification is the successful categorization of these algorithms. To overcome this challenge, the paper introduces a novel hierarchical taxonomy that organizes algorithms into more intricate categories and distinct techniques. The proposed taxonomy is constructed on a four-tier structure, beginning with the broad methodology category, and ending with specific sub-techniques. This structure facilitates a more systematic and exhaustive categorization of algorithms, aiding researchers in grasping the connections between different algorithms and techniques. Included in the paper are both empirical and experimental assessments to classify the various techniques. The empirical assessment judges the techniques according to four standards. The experimental evaluations carry out the following rankings: (1) the algorithms that employ the same specific sub-technique, (2) the different sub-techniques that employ the same technique, (3) the different techniques that employ the same methodology sub-category, and (4) the different methodology sub-categories within the same methodology category. By merging the new methodological taxonomy, empirical analyses, and experimental evaluations, the paper provides a multifaceted and thorough comprehension of the machine learning methods and algorithms for PPI detection. This synthesis helps researchers make well-informed decisions. In its conclusion, the paper furnishes crucial insights into the future possibilities of machine learning techniques for PPI identification, underscoring potential advancements and areas ripe for continued exploration and development

show abstract

Accelerated Transfer Learning for Protein-Protein Interaction Prediction

Cited by 3 publications

References 72 publications

Sequence-based Protein Interaction Site Prediction using Computer Vision and Deep Learning

Sequence-based Protein Interaction Site Prediction using Computer Vision and Deep Learning

Improving the Protein-Protein Interaction Prediction Engine (PIPE) with Protein Physicochemical Properties

Employing Machine Learning Techniques to Detect Protein-Protein Interaction: A Survey, Experimental, and Comparative Evaluations

Contact Info

Product

Resources

About