Turhan, K. | bioRxiv Science

bioRxiv2020

Evaluation of Methods for Protein Representation Learning: A Quantitative Analysis

Data-centric approaches have been utilized to develop predictive methods for elucidating uncharacterized aspects of proteins such as their functions, biophysical properties, subcellular locations and interactions. However, studies indicate that the performance of these methods should be further improved to effectively solve complex problems in biomedicine and biotechnology. A data representation method can be defined as an algorithm that calculates numerical feature vectors for samples in a dataset, to be later used in quantitative modelling tasks. Data representation learning methods do this by training and using a model that employs statistical and machine/deep learning algorithms. These novel methods mostly take inspiration from the data-driven language models that have yielded ground-breaking improvements in the field of natural language processing. Lately, these learned data representations have been applied to the field of protein informatics and have displayed highly promising results in terms of extracting complex traits of proteins regarding sequence-structure-function relations. In this study, we conducted a detailed investigation over protein representation learning methods, by first categorizing and explaining each approach, and then conducting benchmark analyses on; (i) inferring semantic similarities between proteins, (ii) predicting ontology-based protein functions, and (iii) classifying drug target protein families. We examine the advantages and disadvantages of each representation approach over the benchmark results. Finally, we discuss current challenges and suggest future directions. We believe the conclusions of this study will help researchers in applying machine/deep learning-based representation techniques on protein data for various types of predictive tasks. Furthermore, we hope it will demonstrate the potential of machine learning-based data representations for protein science and inspire the development of novel methods/tools to be utilized in the fields of biomedicine and biotechnology.

bioinformatics↗

bioRxiv2019

Personalized Tumor Growth Prediction with Multiscale Tumor Modeling

BackgroundCancer is one of the most complex phenomena in biology and medicine. Extensive attempts have been made to work around this complexity. In this study, we try to take a selective approach; not modeling each particular facet in detail but rather only the pertinent and essential parts of the tumor system are simulated and followed by optimization, revealing specific traits. This leads us to a pellucid personalized model which is noteworthy as it closely approximates existing experimental results.\n\nMethodFor years, research has focused on modeling tumor growth but not many studies have put forward a framework for the personalization of models. In the present study, a hybrid modeling approach which consists of cellular automata for discrete cell state representation and diffusion equations to calculate distribution of relevant substances in the tumor micro-environment is favored. Moreover, naive Bayesian decision making with weighted stochastic equations and a Bayesian network to model the temporal order of mutations is presented. The model is personalized according to the evidence using Markov Chain Monte Carlo. Ultimately, this way of thinking about tumor modeling leads us to a vascular multi-scale model of tumor growth.\n\nResultsTo validate the tumor model, a data set belonging to the A549 cell line is used. The data represents the growth of a tumor for 30 days. We optimize the coefficients of the stochastic decision making equations using first half of the timeline. Then we predict next 15 days of growth without any other supervision. Results are promising with their low error margin and simulated growth data is in line with laboratory results.\n\nConclusionThere are many subsystems which have an effect in the growth of a tumor. A detailed model which includes all of them is currently virtually impossible to implement. We have therefore focused on a system that only includes fundamental components in this study, and have evaluated its predictions. We propose novel probability functions to obtain a personalized model and estimate the individual importance (weights) of each with parameter optimization. Our approach of using simulated annealing for parameter estimation and the subsequent validation of the prediction with in-vitro tumor growth data are, to our knowledge, unique in the literature.

bioinformatics↗