Workshop on Computational Oncology:

Integrating experiments and

computational models

Topics

Dr. Sunaina Banerjee

Yokogawa Technology Solutions India Pvt Ltd, Bangalore

Explainability in AI/ML : design considerations and biological insights

Advanced statistical models can capture complex patterns in a system under exploration. Obtaining a high confidence inference from such models requires a significantly large dataset. In classical machine learning (ML), model reliability is evaluated through cross validation wherein, a dataset is partitioned into training and evaluation. A generalizable model is one in which the model trained on any random partition predicts well on another partition. Training generalizable models, however, requires training data with sufficient statistical variation. In the context of biology, obtaining a large, well-sampled experimental dataset may be limiting due to the challenges of the biological system. This factor limits the application of advanced ML techniques in the biological domain. However, prior knowledge and subject matter expertise can help in an explanation-based model selection for such scenarios. The talk will explore AI explainability tools and how they can be applied to ML applications, especially to domains with restricted data availability. We will illustrate a test case where AI explainability is used for inferring mutational effect on a family of ATP-powered nucleic acid helicases. The talk will further the discussion on building trusted AI systems for biology with the use of AI explainability.

Dr. Ashutosh Srivastava

Biological Sciences and Engineering, IIT Gandhinagar

Understanding molecular basis of circadian rhythms for potential drug discovery

The physiology and behavior of most living organisms is synchronized to the twenty-four hour solar cycle, referred to as circadian rhythms. These rhythms are regulated by complex molecular mechanisms within cells, involving feedback loops and interactions between transcription factors, cryptochromes, and kinases. Clock genes, which regulate these rhythms, play a significant role in the development of various cancers, such as breast, prostate, lung, pancreatic, and ovarian cancers. Mutations in clock genes have been linked to cancer progression, and targeting clock proteins has emerged as a potential therapeutic strategy.

In our previous studies, we have shown the crucial role of structure and dynamics for isoform specific response to small molecules binding to cryptochrome, a core clock protein. In this study, we focus on REV-ERBβ, a nuclear receptor that regulates the transcription of the core clock protein BMAL1. Small molecule agonists and antagonists of REV-ERBs have been identified, but the structural basis of their actions remains unclear due to the lack of crystal structures. Using computational methods, we examined the dynamics of REV-ERBβ's ligand-binding domain (LBD) in different conformations. We found that antagonist binding destabilizes the NCoR peptide's interaction with REV-ERBβ, which could inform the design of more selective and effective small molecules targeting REV-ERBβ.

Dr. Aditya Padhi

Laboratory for Computational Biology & Biomolecular Design, School of Biochemical Engineering, School of Biochemical Engineering, IIT (BHU) Varanasi

Integrating Molecular Dynamics and Computational Modeling for Cancer Nanobody Design and Insights into Immune Dysregulation

Molecular dynamics (MD) simulations and computational modeling have become indispensable tools for advancing our understanding of protein structure-function relationships and their implications in disease. The talk begins by introducing MD simulations and their principles, followed by two case studies that leverage these techniques in cancer- and immunodeficiency-related research. The first case study focuses on enhancing the affinity of nanobodies for critical cancer targets, e.g., the epidermal growth factor receptor (EGFR). We present an integrated computational pipeline that incorporates structure-based protein design, MD simulations, and free energy calculations to predict the binding affinity of non-natural amino acid (nnAA)- incorporated nanobodies. By integrating halogenated tyrosines into nanobodies, we demonstrate a magnitude increase in affinity, supported by experimental validation through surface plasmon resonance (SPR). The second case study explores a novel immune dysregulation mechanism in primary immunodeficiency, caused by a heterozygous IKZF3 missense variant in the human AIOLOS protein. Using MD simulations and computational approaches, we reveal how this variant impairs adaptive immunity by disrupting its DNA-binding function and AIOLOS-IKAROS heterodimerization, leading to defects in B- and T-cell differentiation. These studies highlight the potential of computational modeling and biomolecular simulations in uncovering novel therapeutic strategies and understanding molecular mechanisms of complex diseases like cancer.

Prof. Akhilesh Kumar Singh

Department of Oral and Maxillofacial Surgery, IMS BHU Varanasi

"To be updated soon"

Prof. Ritu Gupta

Laboratory Oncology Unit, AIIMS New Delhi

Computational Oncology: Refining Risk Assessment, Prognostication & Deciphering Oncogenesis

Risk stratification in oncology helps in identifying patients with specific treatment and monitoring requirements. In recent times, data analytics including advanced machine learning (ML) methods have been used to extract valuable information from medical records. Machine learning algorithms have been shown to be useful in devising risk stratification system, both, in benign and malignant disorders.

Identification of the genomic features responsible for the progression of cancer either from its precancerous stage and to a relapsed refractory state can improve the understanding of the disease pathogenesis and, help in devising suitable preventive and treatment measures. However, limited availability of linear longitudinal genomic data from same patients and data handling tools to simultaneously assess the multitude of genomic features and their impact on outcomes are some of the challenges that need to be addressed.

We have developed an innovative AI-based model that can discover pivotal genomic biomarkers that can potentially distinguish premalignant state from overt malignant state. Furthermore, using computational AI tools, we deciphered genomic alterations of prognostic significance that can be used to design genomic testing panels of clinical relevance.

The advanced machine learning algorithms are helpful in delineating the risk conferred by various disease characteristics and can be exploited in developing better risk stratification models. One such model for oncology practice will be discussed. The AI driven algorithms can substantially contribute to the understanding of cancer pathogenesis and are useful in developing esoteric clinical testing algorithms.

In this lecture, I shall discuss the application of machine learning methods for developing ethnicity-based risk stratification systems in oncology using multiple myeloma as a prototype. In addition, the use of artificial intelligence to discover pivotal genomic biomarkers of diagnosis and disease progression will also be discussed.

Dr. Rajnish Kumar

Department of Pharmaceutical Engineering & Technology, IIT (BHU) Varanasi

AI enabled discovery of Choline Acetyltransferase Inhibitors for the treatment of Lung Cancer

Lung cancer cells possess the necessary proteins for acetylcholine (ACh) production, including choline acetyltransferase (ChAT), which catalyzes ACh synthesis. These cells transport ACh via vesicular acetylcholine transporters and organic cation transporters (OCTs), while acetylcholinesterase and butyrylcholinesterase enzymes enable its degradation. The released ACh then binds to nicotinic and muscarinic receptors on lung cancer cells, driving their proliferation, migration, and invasion. This makes ChAT a promising therapeutic target for cancer treatment.

Traditional virtual screening methods struggle with the rapidly growing size of chemical databases, which now exceed billions of compounds. To address this, we employed Deep Docking (DD), a deep neural network-based virtual screening platform, to efficiently screen a 1.3 billion compound library from the ZINC20 database. Through multiple iterations, we refined our hits from an initial 116 million to a final 168,447, ultimately identifying five promising ChAT inhibitors. These inhibitors hold significant promise as therapeutic leads for cancer, offering a targeted approach to treating this deadly disease by inhibiting a key enzyme involved in tumor growth and spread.

Dr. Rakesh Pandey

Dept. of Bioinformatics, BHU, Varanasi

Leveraging Single-Cell RNA Sequencing to Uncover Gene Regulatory Networks Driving Leukemia Therapy Resistance

Targeted cancer therapies have brought transformative changes in patient care, yet resistance to treatment remains a significant challenge, often leading to disease relapse and progression. This presentation delves into how single-cell RNA sequencing (scRNA-seq) is revolutionizing our understanding of therapy resistance in Chronic Myeloid Leukemia (CML) patients by enabling detailed exploration of gene expression patterns at the single-cell level. By constructing gene regulatory networks (GRNs) from scRNA-seq data, we can map the molecular interactions that drive resistance within cancer cell populations. These networks reveal key regulatory factors and pathways along with the cellular mechanisms underlying treatment failure. Through the integration of single-cell data and computational models, this approach could identify potential molecular targets, ultimately guiding new strategies to overcome resistance and improve therapeutic outcomes. This presentation highlights how computational advances combined with experimental data are shaping the future of precision oncology.

Dr. Hem Chandra Jha

The Department of Biosciences and Biomedical Engineering, IIT Indore

Integrating in vitro and in silico approach to study the metabolic pathways and identification of therapeutic targets associated with oral and gastric cancer progression

Indian population has experienced a significant rise in the oral cancer and gastric cancer cases in the past few decades. Our lab works to study pathogenesis and treatment modalities of oral and gastric cancer. The habitual consumption of tobacco and areca nut have been reported to cause oral submucous fibrosis (OSMF) and its malignant transformation into oral squamous cell carcinoma (OSCC). We established OSMF model by treating fibroblast cells with areca nut and observed cellular and molecular changes. We further transferred the conditioned media from these treated cells to the keratinocyte cells and observed the changes associated with OSCC. The keratinocytes cells were also treated directly with areca nut, tobacco and slaked lime to investigate their differential effects on the progression to OSCC. The metabolites and their associated pathways were identified using Liquid Chromatography and Mass Spectrometry (LCMS). We also performed mRNA expression profile for top twenty five genes associated with OSCC. It was observed that there was significant alterations associated with OSCC in the cells treated with tobacco compared to areca nut. Moreover, we found that slaked lime enhances the effects of both tobacco and areca nut and increases their carcinogenicity.

We also investigated the therapeutic potential of FDA-approved kinase inhibitors, targeting kinase that have structural similarity to Aurora kinase A (AURKA), a critical regulator of cell cycle progression, which is frequently overexpressed in gastric and various other cancers. We performed molecular docking and molecular dynamics simulation analyses to screen these inhibitors for their binding affinity to AURKA. Subsequent in vitro experiments demonstrated that the selected kinase inhibitors significantly reduced both the transcript and protein expression levels of AURKA. These inhibitors also decreased ectopic expression of AURKA and downregulated downstream signalling markers in AGS gastric cancer cells. Our findings highlight the ability of kinase inhibitors to modulate inflammatory, proliferative, and apoptotic pathways through AURKA inhibition, suggesting their promise as therapeutic agents in gastric cancer treatment.

Dr. Gowri Balachander

School of Biomedical Engineering, IIT (BHU) Varanasi

From organoids to organ-on-a-chip. Transforming biology in 3D for meaningful pursuits

Three-dimensional (3D) models are increasingly realized as better and improved in vitro models to mimic in vivo-like cellular behavior. Every tissue presents a distinct microenvironment with a unique blend of biochemical and biophysical components that dictate cellular behavior. Recreation of critical features of the tissues that nurtures recapitulation of in vivo-like cellular behavior is the essence of an effective 3D model. In this lecture, through specific examples of 3D models for tissue development and cancer, we will revisit the fundamental principles of designing 3D organoid models, the need and applications of organ-on-a-chip technologies which can effectively recapitulate critical features of the tissues in vitro and applications of such models in mechanistic studies and pre-clinical drug testing.

Dr. Brijesh Kumar

School of Biomedical Engineering, IIT (BHU) Varanasi

Impact of Ethnicity on normal breast and breast cancer biology

Breast cancers are classified into five intrinsic subtypes based on gene expression profile. It is suggested that these intrinsic subtypes originate from a specific developmental stage of breast epithelial cell hierarchy, stem-progenitor-mature cell. However, normal breast epithelial cell lines representing these intrinsic subtypes are yet to be created. Using normal breast tissues of ancestry-mapped Caucasian, African American, and Hispanic women and a primary cell culturing system that allows growth of normal epithelial cells of different developmental stages including estrogen receptor-positive mature luminal cells, we created 15 human telomerase-immortalized breast epithelial cell lines. These cells formed acini on a matrigel and ductal structures on 3-dimensional collagen or hydrogel, indicating that these cell lines have retained characteristics of normal breast epithelial cells. RNA sequencing and PAM50 intrinsic subtype clustering algorithms were used to identify the intrinsic subtypes of the immortalized cell lines together with two well characterized “normal” breast epithelial cell lines MCF10A and HMEC as well as luminal breast cancer cell line MCF-7. Unlike MCF10A and HMEC, which are enriched for basal-like gene expression patterns, our cell lines are classified into luminal A, basal, and normal-like subtypes. This was also reflected in the immunofluorescence staining with basal marker KRT14 and luminal marker KRT19. Few of these cell lines were dual positive for KRT14 and KRT19, but in varying proportions. Cell lines representing claudin-low subtypes were also created, which are phenotypically (CD201+/EpCAM-) different from the above cell lines. Cell lines showed inter-individual differences in stemness/differentiation capabilities and variable basal activity of signaling molecules such as NF-kB, AP-1 and pERK, which is consistent with, possibly reflecting, recent discoveries of genetic variations in gene regulatory regions among general population that contribute to widespread differences in gene expression/signaling under “normal” state. As majority of breast cancers are believed to originate from luminal progenitor cells, which are well represented our cell lines, these cell lines are ideal to delineate the impact of inter-individual and ethnic differences in normal breast biology on breast cancer initiation and progression as well as to determine whether cell-type-origin instead of genomic aberration drives intrinsic subtype-enriched gene expression patterns in breast tumors.

Prof. Neeraj Sharma

School of Biomedical Engineering, IIT (BHU) Varanasi

"To be updated soon"

Dr. Mohit Kumar Jolly

Department of Bioengineering, IISc Bengaluru

What does not kill cancer can make it stronger – Dynamical mechanistic modeling of drug-induced cell-state switching

Drug resistance and consequent tumor relapse are major clinical challenges in cancer treatment. While pre-existing subpopulations of genetically distinct drug-resistant cells have been shown to contribute to this phenomenon, recent studies reveal the role of drug-induced cell-state switching in aggravating disease outcomes by unlocking phenotypic plasticity and heterogeneity. To overcome this challenge, it is essential to understand the emergent dynamics of underlying regulatory networks enabling such switching. I will present several examples from our work integrating computational modeling of regulatory networks, single-cell high-throughput data and experimental validation to better understand what trajectories cells take to evade targeted therapy as well as immunotherapy in different cancers. We present an in silico platform to rationally identify combinatorial therapies to minimize cancer aggressiveness.

Prof. Kartiki Desai

NIBMG Kolkata

Utility and challenges in applying liquid biopsy approaches to cancer detection and progression

Early detection and advances in the first line of cancer therapy have reduced cancer deaths. However, recurring cancers and cancers that spread to other parts of the body remain difficult to treat. While tumor biopsies have greatly advanced our understanding of cancer, second biopsies from longitudinal sampling are rarely available. To understand recurrent disease is challenging and therefore strategies to treat are fewer. is difficult over time. Currently, monitoring these tumors using novel liquid biopsy-based approaches that are minimally invasive, cost-effective and have higher patient compliance is being undertaken. We will discuss design, samples needed, and challenges associated with sampling of RNA/DNA or proteins from bodily fluids and the utility of such data in clinical practice.

Dr. Prasun Kumar Roy

Shiv Nadar Institution of Eminence Deemed to be University, Noida

Additional End-cycle Impulse of Antineoplastic Therapy for Tumour Eradication: Insights from Spontaneous Cancer Regression Phenomenon

Investigation of spontaneous regression phenomenon of earlier lethal diseases have produced landmark therapies, e.g. for syphilis, anaphylaxis and schizophrenia. The natural episodic spontaneous cancer regression (SCR) behaviour is well-documented in humans, e.g. Wisconsin/Scadinavian breast-cancer registries [3.04 million/2.32 lakh cohort-size] respectively show 24%/42% tumours undergo SCR (PMIDs:17032893;21996169). Accordingly, we investigate the natural episodic SCR process which enables host-tissue-mediated permanent tumour elimination, without any toxicity/tumour-relapse that are main drawbacks in oncological treatment. We formulate the cell-kinetics/differential equations of neoplasm/host-immunological interactions (lymphocytes, cytokines). Since malignant cell-kill is first-order kinetics, tumour cell-population decreases exponentially, thus always keeping some neoplastic cells alive asymptotically. For inducing extinction of this population, we formulate negative-bias dynamics. By computational modelling, we identify SCR’s signature behaviour: Occurrence of a second spike of antitumor lymphocyte activation near the end of regression process. This spike is needed for activating lymphocytes that migrate to eliminate residual neoplastic cells sheltered in deep skeletal sanctuary-sites where chemotherapy cannot diffuse. We validate our formulation by: (1)Preclinical animal model: time-varying dynamical analysis of experimental immunohistochemistry/microarray gene-expression mapping across SCR process of melanoma, fibrosarcoma, histiocytoma; (2)Clinical patients: Endogenous regression of osteosarcoma, glioma, epithelioma. Hence, mathematical-computational oncology approaches may furnish unique strategies for anticancer therapy, validated by patient-based applicability.

Dr. Sumanta Mukherjee

Research Scientist, IBM Bengaluru

Representation Learning as a tool to gain insight into biological data

AI-ML is a collection of computational tools that enables automated complex decisionmaking. This domain has evolved from statistical learning theory which assumes partial ordering of the observation space, i.e., it treats a pair of observation either as equal, or orders one as greater than the other. Unlike classical machine learning methods which rely on hand-coded features from domain experts, deep learning models are trained on unstructured data. A deep learning model learns relevant representation (‘implicit features’) to a given task during its training process. The model derives an abstract vector representation that defines a partial order and is aligned to carry out a specific task. Exploratory analysis of these learned vector representations often provides a deeper understanding, which may assist in knowledge discovery.

This talk will introduce the basics of representation learning, and a few necessary computational tools for exploratory high-dimensional analysis. The talk will also cover an representation learning in application for biological sequences, and how they can aid in identification of somatic mutations associated with different histological tumor types.

Dr. Jairam Meena

Department of Pharmaceutical Engineering & Technology, IIT (BHU) Varanasi

Integration of Neoantigen epitopes with Artificial Intelligence inspired multi-epitopes for personalized cancer vaccine design

Current work focused on the design of multi-epitope peptide vaccine for hepatocellular carcinoma through computational approaches. Potential B and T cell epitopes were against the melanoma protein. Epitopes were selected based on their strong IFN-g stimulation and limited allergenic response. Identified epitopes were integrated with the reported neoantigen epitopes for hepatocellular carcinoma. β-defensin 1 was added as an adjuvant, and linkers were employed to link the epitopes and adjuvants in the final vaccine. The vaccine showed a high antigenicity score, was non-allergenic, non-toxic, and demonstrated stable physicochemical properties, including good solubility. Docking results indicated a ~14.5 kcal/mol binding affinity for the TLR4-vaccine interaction. MD simulations confirmed stability, with RMSD and hydrogen bond counts indicating consistent structure. Immune simulations showed robust responses, including memory formation in B and T cells, with a shift from IgM to IgG1/IgG2 dominance. Cytokine analysis revealed peak IFN-γ levels post-exposure, highlighting the vaccine’s strong immune-stimulating potential.

Prof. Shubhra Ghosh Dastidar

Bose Institute, EN 80, Sector V, Bidhannagar, Kolkata, India

Allosteric Changes in Cancer Drug Targets: Insights from Jiggling Ball-and-Stick Models

Allosteric regulations of the functions of the biological macromolecules are of tremendous importance. Allostery could also be induced by administering small molecules and that possibility gives enormous advantage in the of drug design initiatives. While state-of-art experiments can suggest the existence of allosteric transitions in a molecular system, the mechanisms and atom-atom interaction network of allostery in a molecule are often non-trivial to understand just from the static models of molecular structures. The necessary extra miles of insights are obtainable from the computational modelling and simulations of their dynamics which can show the jiggling atoms in action, revealing a more realistic scenario. But how much it can really help to take the scientific understanding of allostery forward for possible applications? This presentation will address this issue with examples of various molecular systems like α,β-Tubulin, Bcl2, kinases, all of which are of tremendous importance in the field of drug design to combat cancer. Not only the structural and thermodynamic insights which will be presented, but also our latest initiatives to use machine learning methods will be briefly discussed.

For further inquiries: wocon2024@itbhu.ac.in

Page updated

Report abuse