Publication: Machine learning approaches in glioma to identify subtype-specific candidate biomarkers and therapeutics
Abstract
Yaygın görülen bir beyin tümörü olan glioma, hastalığın karmaşık ve agresif seyri nedeniyle tanı, prognoz ve tedavi açısından önemli klinik zorluklar sergilemektedir. Bu özellikler doğru tanı, prognoz ve etkili tedavinin geliştirilmesinde zorluklara yol açmaktadır. Mevcut moleküler belirteçlerin sınırlamaları ve etkili tedavi seçeneklerinin eksikliği, kişiselleştirilmiş tedavi stratejilerini geliştirmek için sağlam, alt türe özgü sistem biyobelirteçlerinin ve yeni ilaç adaylarının tanımlanmasına yönelik acil ihtiyacı vurgulamaktadır. Bu çalışma, gen ekspresyon verilerini kullanan bütünleştirici bir makine öğrenimi ve biyoinformatik boru hattı aracılığıyla glioblastoma (GBM) ve düşük dereceli glioma (LGG) tümörleri için ayırt edici sistem biyobelirteçlerini ve yeni terapötik ajanları belirlemeye odaklanmıştır. Potansiyel ilaç hedefleri olarak alt türe özgü anahtar genleri belirlemek üzere aşırı örnekleme ve özellik seçimi için çeşitli makine öğrenimi algoritmaları kullanılmıştır. Bu yaklaşım, yüksek tanısal yetkinlik ve dolayısıyla potansiyel klinik uygunluk gösteren bir gen kümesinin tanımlanmasını sağlamaktadır. Glioma teşhisinde kullanılma potansiyeli olan toplam 11 anahtar gen tanımlanmıştır: GBM için AB11FIP4, TYRO3, THEM5, SST, SMIM32, MIGA1, ARFGEF3 ve ANK3; LGG için ise GUCA1A ve CES4A. Bu genlerin, kişiselleştirilmiş tedavi amacıyla ilaç hedefi olarak işlev görme kapasitesi bulunmaktadır. İlaç-gen etkileşimlerine dayanan ilaç yeniden konumlandırma (drug repurposing) analizleri sonucunda, GBM için altı onaylı ilaç belirlenmiştir. Bunlar arasında iki antineoplastik ajan (vandetanib ve capecitabine) ve dört sinir sistemi ajanı (melatonin, agomelatine, ramelteon ve tasimelteon) bulunmaktadır. Buna ek olarak, mukolitik bir ajan olan ambroksol, LGG için potansiyel bir tedavi olarak tanımlanmıştır. Bu bulgular, bu ilaç adaylarının bağlanma yeteneklerini doğrulayan moleküler kenetlenme simülasyonları ile desteklenmiştir. Bu çalışma, sistem biyobelirteçlerinin tanımlanması ve yeni ilaçların eldesi için makine öğrenimi, sistem biyolojisi ve yapısal biyolojinin entegre edilmesinin umut verici geleceğini göstermektedir. Önerilen terapötik hedefler ve bileşikler, ileri deneysel çalışmalarla doğrulanmayı hak eden cesaret verici sonuçlar ortaya koyarak, glioma tedavisinde kişiselleştirilmiş tedavi stratejilerinin geliştirilmesine katkı sağlayabilecek niteliktedir.
Glioma, a prevalent brain tumour, demonstrates significant clinical challenges in terms of diagnosis, prognosis, and treatment due to the intricate and aggressive course of the disease. These characteristics give rise to challenges in accurate diagnosis, prognosis, and the development of effective treatment. The limitations of current molecular markers and the lack of effective therapeutic options highlight the urgent need for the identification of robust, subtype-specific systems biomarkers and novel drug candidates to improve personalised treatment strategies. The present study focused on identifying distinctive systems biomarkers and novel therapeutic agents for glioblastoma (GBM) and low-grade glioma (LGG) tumours through an integrative machine learning and bioinformatics pipeline using gene expression data. Several machine learning algorithms were utilised for oversampling and feature selection to identify subtype-specific key genes as potential drug targets. This approach enabled the identification of a set of genes showing high diagnostic capability, thus potential clinical relevance. A total of 11 key genes were identified as having potential for use in the diagnosis of glioma: AB11FIP4, TYRO3, THEM5, SST, SMIM32, MIGA1, ARFGEF3, ANK3 for GBM and GUCA1A and CES4A for LGG. These genes have the capacity to function as drug targets for the advent of personalised medicine in the future. The subsequent drug repurposing analysis, which was based on drug-gene interactions, identified six approved drugs for GBM. These included two antineoplastic agents (vandetanib and capecitabine), and four nervous system drugs (melatonin, agomelatine, ramelteon, and tasimelteon). In addition, ambroxol, a mucolytic agent, was identified as a potential treatment for LGG. These findings were further corroborated by molecular docking simulations, which confirmed the binding capabilities of these drug candidates. The present study demonstrates the promising future of integrating machine learning, systems biology, and structural biology for the recognition of systems biomarkers and the acquisition of new drugs. The proposed therapeutic targets and compounds have yielded encouraging results that merit further validation in through experimental studies, paving the way for the development of personalised therapeutic strategies in the treatment of glioma.
Glioma, a prevalent brain tumour, demonstrates significant clinical challenges in terms of diagnosis, prognosis, and treatment due to the intricate and aggressive course of the disease. These characteristics give rise to challenges in accurate diagnosis, prognosis, and the development of effective treatment. The limitations of current molecular markers and the lack of effective therapeutic options highlight the urgent need for the identification of robust, subtype-specific systems biomarkers and novel drug candidates to improve personalised treatment strategies. The present study focused on identifying distinctive systems biomarkers and novel therapeutic agents for glioblastoma (GBM) and low-grade glioma (LGG) tumours through an integrative machine learning and bioinformatics pipeline using gene expression data. Several machine learning algorithms were utilised for oversampling and feature selection to identify subtype-specific key genes as potential drug targets. This approach enabled the identification of a set of genes showing high diagnostic capability, thus potential clinical relevance. A total of 11 key genes were identified as having potential for use in the diagnosis of glioma: AB11FIP4, TYRO3, THEM5, SST, SMIM32, MIGA1, ARFGEF3, ANK3 for GBM and GUCA1A and CES4A for LGG. These genes have the capacity to function as drug targets for the advent of personalised medicine in the future. The subsequent drug repurposing analysis, which was based on drug-gene interactions, identified six approved drugs for GBM. These included two antineoplastic agents (vandetanib and capecitabine), and four nervous system drugs (melatonin, agomelatine, ramelteon, and tasimelteon). In addition, ambroxol, a mucolytic agent, was identified as a potential treatment for LGG. These findings were further corroborated by molecular docking simulations, which confirmed the binding capabilities of these drug candidates. The present study demonstrates the promising future of integrating machine learning, systems biology, and structural biology for the recognition of systems biomarkers and the acquisition of new drugs. The proposed therapeutic targets and compounds have yielded encouraging results that merit further validation in through experimental studies, paving the way for the development of personalised therapeutic strategies in the treatment of glioma.