Page 56 - TD-4-1

Tumor Discovery Drug repurposing for pancreatic cancer via AI

1. Introduction trials. Given the high failure rate in clinical trials, there
12
is a need for a more efficient drug development system.
Pancreatic ductal adenocarcinoma (PDAC), often referred Recent advancements in deep learning have shown
to as the “king of cancers,” is the most common type of promise in drug discovery, with neural network methods
pancreatic cancer. It is primarily characterized by a being applied to predict drug-target interactions (DTI). 13-16
1
lack of significant symptoms in its early stages, making DTI data from various databases can help elucidate the
timely diagnosis challenging. Even when symptoms do relationships between drugs and their targets. 13,14 By
emerge, they are frequently mistaken for other health framing drug and target features as a binary classification
conditions, typically indicating an advanced stage of the problem, DTI models based on deep neural networks
disease. Scholars predict that by 2030, pancreatic cancer (DNNs) can predict the interactions between drugs and
will become the second leading cause of cancer-related targets (biomarkers), thereby identifying candidate multi-
2
deaths, resulting in an estimated 46,000 deaths annually molecular drugs for specific diseases. 15,16 This approach,
by 2040. Despite numerous technological advancements known as drug repurposing, utilizes existing drugs for
3
in cancer treatment, the 5-year survival rate for PDAC new therapeutic purposes, potentially expediting the
patients remains only 12.2%. A major contributing factor progression to preclinical and clinical trials. 17
4
to this low survival rate is that pancreatic cancer is often
diagnosed at an advanced stage, which severely limits the Before employing the DNN-based DTI model to predict
opportunities for surgical intervention. Compared to other potential multi-molecular drugs for PDAC biomarkers, we
cancers, pancreatic cancer has a relatively high mortality adopted a systems biology approach using whole-genome
rate, influenced by various factors such as geographic microarray data from PDAC and health controls to study
region, case numbers, and medical standards. The overall their genome-wide genetic and epigenetic networks
5
survival rate for PDCA is relatively low, primarily because (GWGENs) for investigating the carcinogenic mechanisms
tumors are often diagnosed at advanced stages and due to of PDAC. The first step involved mining large databases
factors such as tumor location, size, and metastasis that to establish candidate GWGENs. Next, using system
affect treatment efficacy. Specific survival rate data may identification and systematic order detection techniques
vary by region and treatment modality, reflecting the on the corresponding genome-wide microarray data of
varying prevalence of pancreatic cancer across different PDAC and healthy controls to eliminate false positive
populations. This variability is also influenced by various interactions and regulations, we obtained real GWGENs
risk factors, including smoking, high-fat diets, obesity, and for PDAC and healthy controls. Given that the Kyoto
6
genetic factors. Therefore, there is a pressing global need Encyclopedia of Genes and Genomes (KEGG) pathway
for enhanced preventive measures and genetic diagnostics annotations currently encompass only up to 6,000 nodes
to address this significant health challenge. 7 of GWGENs, the principal network projection (PNP)
method was employed to individually extract the top 6,000
The carcinogenic mechanisms underlying PDAC involve important nodes as core GWGENs from the real GWGENs
multiple gene mutations, abnormalities in signaling pathways, of both PDAC and healthy controls. Then, we annotated
and the influence of the tumor microenvironment. Common the real GWGENs using KEGG pathways to construct core
gene mutations in PDAC include KRAS, CDKN2A, TP53, signaling pathways for both PDAC and healthy controls. By
and SMAD4. These mutations disrupt signaling pathways, comparing the similarities and differences between the core
8
leading to uncontrolled cell proliferation, inhibition of signaling pathways of PDAC and healthy controls, along
apoptosis, and promotion of tumor growth and metastasis.
with their downstream cellular functional abnormalities,
Current treatment options for pancreatic cancer include we investigated the carcinogenic mechanisms of PDAC
surgical resection, radiation therapy, chemotherapy, and identified significant biomarkers, including c-MYC,
immunotherapy, and targeted drugs. However, not all forkhead box O3 (FOXO3), and tumor suppressor p53
patients are suitable for surgery. Common chemotherapy (TP53), as potential drug targets. Finally, we combined the
9
regimens often fail due to drug resistance, resulting from features of selected biomarkers with those of molecular
complex interactions among pancreatic cells, cancer cells, drugs to train the DNN-based DTI model, predicting the
and the tumor microenvironment. Consequently, clinical probability of interaction between candidate molecular
10
outcomes remain poor, and effective treatment methods drugs and the identified drug targets (biomarkers).
are still lacking. Moreover, drug development typically Based on drug design specifications, such as regulatory
takes at least 10 years and requires substantial funding. capacity, sensitivity, and toxicity, we identified potential
11
Biopharmaceutical companies must invest heavily in drugs, ultimately selecting the combination of MK-2206
research to identify drug targets, assess efficacy, consider and gemcitabine as a promising multi-molecular drug
side effects, and conduct extensive preclinical and clinical approach to target key biomarkers for PDAC treatment.

Volume 4 Issue 1 (2025) 48 doi: 10.36922/td.4709