Machine learning in precision medicine of cancer treatment

Размещено на http://www.allbest.ru/

Faculty computer science and software engineering, international information technology university

Machine learning in precision medicine of cancer treatment

Kokenova U.K.

Almaty, republic of Kazakhstan

Abstract

Cancer is genetic illness induced by bodily mutations in the genomes. The distinction among “driver” and “passenger” mutation is a important task for the functionalization of cancer genomics in the treatment of patients. the assignment is to distinguish the mutations that conduct to growth of mutational nucleus (drivers) from the benign (passenger) mutations. Typically, the search for definitions and analysis of these pathogenic cells is done manually. We need to automate the analysis of text and machine learning algorithms. Also we should compare different methods to find the best one. The minimum value for logg-loss is obtained when applying k nearest neighbour algorithm. Keywords: naive Bayes, k-nearest neighbour (kNN), logistic regression (LR), linear support vectors (linear SVM), random forest. mutation genomics benign cancer

Аннотация

МАШИННОЕ ОБУЧЕНИЕ В ТОЧНОЙ МЕДИЦИНЕ ЛЕЧЕНИЯ РАКА

Кокенова У.К. (Республика Казахстан)

Кокенова Улжан Кажимукановна - магистр, факультет вычислительной техники и программного обеспечения, Международный университет информационных технологий, г. Алматы, Республика Казахстан

Рак - это генетическое заболевание, вызванное телесными мутациями в геномах. Различие между «водительской» и «пассажирской» мутацией является важной задачей для функционализации геномики рака при лечении пациентов. Задача состоит в том, чтобы отличить мутации, которые ведут к росту мутационного ядра (драйверов) от доброкачественных (пассажирских) мутаций. Как правило, поиск определений и анализ этих патогенных клеток производится вручную. Нам нужно автоматизировать анализ текста и алгоритмы машинного обучения. Также мы должны сравнить различные методы, чтобы найти лучший. Минимальное значение Logg-Loss получается при применении алгоритма kNN.

Ключевые слова: наивный Байес, метод k-ближайших соседей (kNN), логистическая регрессия (LR), линейные опорные вектора (linear SVM), случайный выбор.

Cancer is known as most frequent illness facing humanity. It is known that above than 200 types of cancer exist. Also specific molecular patterns that require unique therapeutic approaches can be described for each form. Also that illness includes complex changes in the gene [1]. In each type of cancer, architecture of recurring genetic aberrations like somatic mutations, changes in copy numbers, altered gene expression profiles and various epigenetic modifications is special. Need for improved cancer detection, care and preclusion has grown, and is closely connected with a greater understanding of tumor genetic modifications.

Overview. In the last years was a lot of discussion at how precision medicine and, more specifically, how genetic testing could affect the treatment of diseases like cancer. Yet its merely happens fractionally due to of the enormous quantum of manual work that is still needed. Sequencing is a technique for estimating the casualties in the DNA that triggers the body's inherited disorders, inherited predispositions or characteristics [2].If a cancer sequenced it can be able to have a lot of mutations. However the assignment is to distinguish the mutations that conduct to growth of mutational nucleus (drivers) from the benign (passenger) mutations. Currently analysis of mutations is satisfied manually. The molecular pathologist chooses a list of important genetic variants which he / she would like to study. In the medical literature the molecular pathologist searches for evidence that is perhaps linked to important genetic variations. We should to change stage with a machine learning. The molecular pathologist would also have to determine which variants are of concern, as well as collect relevant data.

Hence the task is to create an algorithm which can be able to automatically classify using an expert-annotated knowledge base a different genetic variants as a benchmark.

Goal. The aim is to form a machine learning algorithmic program that may mechanically determine genetic variants mistreatment the knowledge domain as a basis. An associate algorithmic program for classifying genetic mutations should be established supported clinical information (text). It might be key to finding more practical therapies for cancer. Biologists could use the list of mutations to develop a private medical care that targets cancer cells of the patient. it's not an easy task, however even for consultants, understanding clinical information is extremely tough. Clinical information modeling (text) would thus be crucial to the success of our strategy. Our essential objective of classifying clinically unjust genetic mutations is thus to cut back and eliminate the manual efforts created by the clinical medical specialist in shaping and cross-referencing the categories of genes from the literature of science.

Usually problems of this types calls precision medicine. A precision medicine or an individual medicine cures diseases by carefully taking into account the genetic characteristics, lifestyle and environment of each patient [4].

We will come up with algorithms during this competition to determine the genetic variants supported by clinical knowledge (text). Genetic variations can also be divided into 9 separate groups. Modeling of scientific facts (text) will be critical to the effectiveness of the strategy.

The Sloan Charles Kettering Memorial Center collects an elite dataset. This particular set of knowledge was elite as a result of this set of knowledge was developed precisely for this reason. This problem was initiated by the Memorial Cancer Center. Sloan Charles Kettering (MSKCC), approved by the NIPS Calling Council 2017 [3].

When constructing a model for a medical problem mistake, it is important to decrease the mistake variable and respond to the forecast in terms of likelihood as specified in the problem definition to eliminate uncertainty and to achieve the most probable type. When forecasting the patient's class, we can also add the explanation for explaining the reasons for our prediction to the Class outcome. Therefore, an apparent part of this forecast will be interpretability.

To do this, it was agreed to use some of the easily interpretable algorithms like Naive Bayes, Logistic Regression, and Linear Support Vector Machine. It has also been attempted to get strong forecasts from less interpretable models including Random Forest and K- Nearest NeighBor. During issue analysis it was discovered that for this forecast the latency can be weakened to some degree without increasing the error rate.

First it is a classification problem because output is a discrete data. And there is exist possibility of multiple discrete output so it is multi class classification problem. When constructing a model for a medical problem mistake, it is important to decrease the mistake variable and respond to the forecast in terms of likelihood as specified in the problem definition to eliminate uncertainty and to achieve the most probable type. General result taken from first stage of algorithm usage shown in Table l.The minimum value for log-loss is obtained when applying kNN algorithm on the value points of the genome, variance, and text data. Although the logg-loss is less for the kNN method, we prefer Logistic Regression (with balancing) since kNN provides an overfitted value (1.008)

Table 1. Results taken from each type of algorithm

References / Список литературы

1. Tomczak K., Czerwinska P. & Wiznerowicz M, 2015. The Cancer Genome Atlas (TCGA): an immeasurable source of knowledge. Contemporary oncology. 19 (1A). A68.

2. Pareek C.S., Smoczynski R. & Tretyn A., 2011. Sequencing technologies and genome sequencing. Journal of applied genetics. 52 (4), 413-435.

3. Wang L., 2017. Proposal of Kaggle's Personalized Medicine Competition, viewed 11

4. Lyu B., Haque A., 2018. Deep learning based tumor type classification using gene expression data.

Размещено на Allbest.ru