Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/171603
Full metadata record
DC FieldValueLanguage
dc.contributor.advisorKusuma, Wisnu Ananta
dc.contributor.advisorHaryanto, Toto
dc.contributor.authorAdhiva, Jeni
dc.date.accessioned2025-11-28T06:32:57Z
dc.date.available2025-11-28T06:32:57Z
dc.date.issued2025
dc.identifier.urihttp://repository.ipb.ac.id/handle/123456789/171603
dc.description.abstractKanker masih menjadi penyebab utama kematian dan menuntut pendekatan terapi yang lebih presisi serta efisien. Peptida terapeutik menawarkan afinitas dan spesifisitas tinggi terhadap reseptor sel kanker. Namun, penemuan kandidat peptida baru melalui uji laboratorium memerlukan waktu dan biaya besar. Penelitian ini mengembangkan pendekatan in silico berbasis deep learning untuk memprediksi interaksi peptida dan protein yang terkait dengan kanker, dengan fokus pada peptide alami dari bisa ular Calloselasma rhodostoma. Tujuan penelitian adalah merancang pipeline prediksi yang akurat dan hemat biaya melalui dua strategi. Pertama, melakukan transfer learning pada model CAMP yang telah pre-trained, kemudian melakukan fine-tuning menggunakan data domain khusus berupa pasangan peptida bisa ular dengan protein terkait kanker. Kedua, menyusun pemodelan berbasis fitur embedding ProtT5 dengan membangun arsitektur SAE-DNN dan TabNet dari representasi satu dimensi pasangan peptida dan protein. Pelatihan menerapkan k=5 fold cross-validation, pencarian hyperparameter optimal, dan evaluasi metrik seperti akurasi, precision, recall, F1 score, dan ROC-AUC. Hasil penelitian menunjukkan bahwa SAE-DNN yang menggunakan representasi fitur ProtT5 memberikan kinerja terbaik dengan akurasi 0,7749 dan ROC-AUC 0,8538, dan dijadikan model utama dalam penelitian ini. Relevansi biologis dievaluasi melalui analisis pengayaan (GO/KEGG) pada prediksi teratas, yang menyoroti sejumlah target utama protein terkait kanker, antara lain TRBC2, CD274, HIF1AN, PCSK9, dan PLAU, yang berperan pada jalur kritis seperti imunomodulasi/immune checkpoint, regulasi hipoksia, metabolisme lipid, dan metastasis. Temuan ini menyajikan prioritas pasangan interaksi peptida dan protein untuk diuji secara eksperimental serta menunjukkan potensi peptida bisa ular sebagai sumber kandidat obat antikanker. Secara keseluruhan, pendekatan yang diusulkan mempercepat penyaringan kandidat dengan efisiensi waktu dan biaya, sekaligus memberikan dasar hipotesis yang kuat untuk validasi lanjutan. Penelitian ini berkontribusi pada pengembangan metode komputasi untuk penemuan peptida antikanker berbasis sumber alami.
dc.description.abstractCancer remains one of the leading causes of mortality, requiring more precise and efficient therapeutic strategies. Therapeutic peptides offer high affinity and specificity toward cellular receptors. However, the discovery of new candidates through laboratory assays is both time-consuming and costly. This thesis presents an in silico deep learning-based approach to predict peptide–protein interactions relevant to cancer, focusing on natural peptides derived from the venom of Calloselasma rhodostoma. The objective of this research is to develop an accurate and cost-effective prediction pipeline by employing two complementary strategies. First, transfer learning was applied to a pre-trained CAMP model, followed by fine-tuning with domain-specific data comprising venom peptide–cancer protein interaction pairs. Second, ProtT5 feature-based embeddings were utilized to construct predictive architectures using a stacked autoencoder–deep neural network (SAE-DNN) and TabNet. Model training involved five-fold cross-validation, optimal hyperparameter tuning, and comprehensive evaluation metrics, including accuracy, precision, recall, F1-score, and ROC-AUC. The results showed that the SAE-DNN model with ProtT5 feature representations achieved the best performance, with an accuracy of 0,7749 and a ROC-AUC of 0.8538, and was therefore selected as the primary model in this study. The biological relevance of the top-ranked predictions was further assessed through enrichment analyses (GO/KEGG), which identified several cancer-related protein targets, including TRBC2, CD274, HIF1AN, PCSK9, and PLAU. These proteins are involved in key oncogenic pathways, including immunomodulation and immune checkpoint signaling, hypoxia regulation, lipid metabolism, and metastasis. Overall, the findings provide prioritized peptide–protein interaction pairs for subsequent experimental validation and highlight the potential of snake venom derived peptides as a promising source of anticancer drug candidates. The proposed approach accelerates the screening process by reducing both time and cost, while offering a robust framework for generating hypotheses for downstream validation. This study contributes to the advancement of computational methodologies for discovering anticancer peptides from natural sources.
dc.description.sponsorshipBeasiswa Indonesia Bangkit (BIB) dari Kementerian Agama Republik Indonesia.
dc.language.isoid
dc.publisherIPB Universityid
dc.titlePendekatan Transfer Learning untuk Prediksi Interaksi Peptida Bisa Ular Calloselasma Rhodostoma dengan Protein Terkait Kanker.id
dc.title.alternativeA Transfer Learning Approach for Predicting Interactions between Calloselasma Rhodostoma Snake Venom Peptides and Cancer-associated Proteins.
dc.typeTesis
dc.subject.keywordbisa ularid
dc.subject.keyworddeep learningid
dc.subject.keywordinteraksi peptida dan proteinid
dc.subject.keywordkankerid
dc.subject.keywordtransfer learningid
Appears in Collections:MT - School of Data Science, Mathematic and Informatics

Files in This Item:
File Description SizeFormat 
cover_G6501231023_1c0f1620429749a5898b549228ba23ed.pdfCover500.55 kBAdobe PDFView/Open
fulltext_G6501231023_15d2907547f54380a1c886334231f1c3.pdf
  Restricted Access
Fulltext2.81 MBAdobe PDFView/Open
lampiran_G6501231023_98da9ab1c993404ebc66936231a51b51.pdf
  Restricted Access
Lampiran234.89 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.