Discovery of Biomarkers by Analyzing Omics Data Wıth Machine Learning Using Biological Domain Knowledge
Discovery of Biomarkers by Analyzing Omics Data Wıth Machine Learning Using Biological Domain Knowledge
Yükleniyor...
Özet
Yüksek verimli omik teknolojilerinin hızla gelişmesi, büyük miktarda biyolojik veri üreterek biyobelirteç keşfi için yeni fırsatlar yaratmıştır. Ancak transkriptomik, metagenomik ve proteomik veri setlerinin yüksek boyutluluğu ve karmaşıklığı, altta yatan biyolojik ilişkileri göz ardı eden geleneksel öznitelik seçim yöntemleri için önemli zorluklar oluşturmaktadır. Bu amaçla, transkriptomik veri analizi için biyolojik alan bilgisi olarak Gen Ontolojisi (GO) terimlerini bünyesinde barındıran, G-S-M tabanlı bir makine öğrenmesi modeli olan GeNetOntology'yi geliştirdik. GeNetOntology, hastalıkla ilişkili gen ontolojilerini başarıyla tespit etmiş ve farklı transkriptomik veriler üzerindeki tahmin modellerinin yorumlanabilirliğini artırmıştır. Bunun yanı sıra, kolorektal kanserin (KRK) metagenomik analizi için Enzim Komisyonu (EC) isimlendirme sistemine dayalı bir G-S-M yaklaşımı önerdik. Sekiz farklı ülkedeki KRK kohortlarından elde edilen topluluk düzeyindeki enzim bolluk profilleri, biyolojik alan bilgisi olarak EC kategorilerine göre gruplandırdık. KRK ile ilişkili çeşitli enzimleri tespit ederek, bunların KRK patogenezindeki potansiyel rollerini belirledik. Son olarak, yüksek dereceli seröz yumurtalık kanserinde kemoterapi yanıtını belirleyen biyobelirteçleri bulmak amacıyla G-S-M makine öğrenmesi yaklaşımını çoklu-omik entegrasyonuna genişlettik. Bu amaçla geliştirdiğimiz integrOmeGSM yaklaşımında, transkriptomik ve proteomik öznitelikleri, protein-transkript ilişkilerine göre gruplandırıldık ve tahmin edici uygunlukları açısından puanlandık. Ayrıca transcriptomik ve proteomik verileri daha önce geliştirdiğimiz biyolojik alan bilgisini kullanan makine öğrenmesi modelleri ile analiz ederek performans karşılaştırması yaptık ve kemoterapi yanıtını belirleyen GO terimlerini belirledik.
The rapid growth of high-throughput omics technologies has generated large amounts of biological data, creating new opportunities for candidate biomarker discovery. However, the high dimensionality and complexity of transcriptomic, metagenomic, proteomic datasets present significant challenges for traditional feature selection methods, which often neglect underlying biological relationships. For this aim, we developed GeNetOntology, which is a G-S-M based machine learning (ML) model incorporating Gene Ontology (GO) terms as external biological domain knowledge for transcriptomic data analysis. GeNetOntology successfully identified disease related gene ontology terms and improved the interpretability of predictive models on different transcriptomics data. Beside that, we proposed an enzyme commission (EC) nomenclature-based G-S-M approach for metagenomic analysis of colorectal cancer (CRC). Community-level enzyme abundance profiles from CRC cohorts were grouped according to their EC categories as biological domain knowledge. The proposed framework identified several CRC-associated enzymes, highlighting their potential roles in CRC pathogenesis. Finally, we extended the G-S-M machine learning approach to multi-omics integration to identify chemotherapy response biomarkers in high-grade serous ovarian cancer (HGSC). In the developed integrOmeGSM approach, transcriptomic and proteomic data were grouped based on shared protein/transcript associations, scored for predictive relevance. We also analyzed transcriptomic and proteomic data using G-S-M based ML models, which utilize biological domain knowledge, to compare model performance. We identified the GO terms and KEGG pathways that determine chemotherapy response in ovarian cancer.
The rapid growth of high-throughput omics technologies has generated large amounts of biological data, creating new opportunities for candidate biomarker discovery. However, the high dimensionality and complexity of transcriptomic, metagenomic, proteomic datasets present significant challenges for traditional feature selection methods, which often neglect underlying biological relationships. For this aim, we developed GeNetOntology, which is a G-S-M based machine learning (ML) model incorporating Gene Ontology (GO) terms as external biological domain knowledge for transcriptomic data analysis. GeNetOntology successfully identified disease related gene ontology terms and improved the interpretability of predictive models on different transcriptomics data. Beside that, we proposed an enzyme commission (EC) nomenclature-based G-S-M approach for metagenomic analysis of colorectal cancer (CRC). Community-level enzyme abundance profiles from CRC cohorts were grouped according to their EC categories as biological domain knowledge. The proposed framework identified several CRC-associated enzymes, highlighting their potential roles in CRC pathogenesis. Finally, we extended the G-S-M machine learning approach to multi-omics integration to identify chemotherapy response biomarkers in high-grade serous ovarian cancer (HGSC). In the developed integrOmeGSM approach, transcriptomic and proteomic data were grouped based on shared protein/transcript associations, scored for predictive relevance. We also analyzed transcriptomic and proteomic data using G-S-M based ML models, which utilize biological domain knowledge, to compare model performance. We identified the GO terms and KEGG pathways that determine chemotherapy response in ovarian cancer.
Açıklama
Anahtar kelimeler
Biyomühendislik, Biyoinformatik, Gastrointestinal Mikrobiyom, Moleküler Biyoloji, Moleküler Genetik, Bioinformatics, Gastrointestinal Microbiome, Molecular Biology, Molecular Genetic
Bilim Dalları
Alıntı
WoS Q
Scopus Q
Kaynak
Cilt
Sayı
Başlangıç Sayfası
Bitiş Sayfası
Koleksiyonlar
Yükleniyor...
