Lecture 20 - AI, Bioinformatics and 'omics approaches to cancer biology

What can Next Generation Sequencing be used for?

• DNA 

  • Whole genome sequencing 

    • Point mutations 

    • Structural rearrangements 

• RNA 

  • RNA Seq 

• Chromatin 

What can next gen sequencing identify?

• Copy number changes 

• Single Nucleotide Variants (SNV) 

• Insertion-deletions (Indels)

• Structural variants  

• Allows comparison of tumour DNA and germline DNA 

How has genome sequencing aided therapeutic interventions?

• Sequencing showed average cancer genomes have 4-5 driver mutations including drivers in non-coding DNA 

  • Many common drivers across cancer types

  • Allows for certain therapies to be used in many different cancers

• No drivers identified in 5% of cancer cases  

How can next generation sequencing aid in preventing cancer?

• Sequencing can show causes of cancer development 

  • Refined and catalogued dozens of mutational signatures  

    • E.g. UV light, tobacco, defective DNA damage repair 

What has next generation sequencing shown us about cancer evolution?

• Certain mutations like TP53 and KRAS tend to occur earlier  

• Certain mutational signatures occur in early clonal stages 

• Mutational burden and genome duplications can be converted to chronological time to estimate latency  

  • Mutations can occur years or decades before cancer diagnosis 

How can next gen sequencing aid in understanding early cancer development?

• Cancer evolutionary history can be reconstructed from sequencing data 

  • Number of reads with mutation + number of copies of gene 

  • Has shown certain mutations like TP53 and KRAS tend to occur earlier

What is expression profiling?

High-throughput molecular technique which can measure the activity of thousands of genes by quantifying mRNA levels

• Expression profiling can be used to experimentally identify biological functions 

  • E.g. Knockdown gene using shRNA then expression profile  

What can expression profiling be used for in the context of cancer?

• Can be used to discover new tumour or tumour sub-groups 

• Can be used to define known tumours or tumour sub-groups 

• Can be used to find expression biomarkers or signatures which predict survival 

• Can be used to find expression biomarkers or signatures which predict response to particular therapies  

How can expression profiles be used to define poor breast cancer survival signature?

• Used Hu25K Agilent 2-colour oligonucleotide array 

• 70 genes chosen by supervised analysis to discriminate patients by prognosis 

  • Good prognosis - >5 years no metastasis 

  • Bad prognosis – metastasis within 5 years 

• Similar method was then used to separate ~300 patients into low-risk group and high-risk group 

  • Allows estimate of survival chance based on identified biomarkers/mutations 

What was the MINDACT trial?

Also called the Mammaprint trial.

• Trail aimed at confirming that the separation of breast cancer patients into ‘low’ risk and ‘high’ risk using genomic testing was effective

• Could ‘low’ risk patients be safely spared chemotherapy without affecting survival

Why was confirming ‘low risk’ patients during the MINDACT trial could be spared chemotherapy a key goal?

• Allows patients to be given less aggressive treatment to prevent as many harmful side effects  

  • Will also help reduce issues following cancer remission as a result of cancer therapy  

How does single cell sequencing work?

• Places cells inside oil droplets  

• Inside oil droplets, the DNA is cleaved and library is created 

• However each DNA fragment is labelled with cell specific barcode which allows single cells to be sequenced and identified among thousands or millions 

What can single cell sequencing be used for?

• Can be used to analyse tumour immune microenvironment 

  • Can indicate sensitivity to immune therapies such as immune checkpoint inhibition  

What are functional genomic screens?

High-throughput experimental approaches which often use CRISPR-Cas9.

• Designed to systematically knockout target genes

• Allows identification of gene function and link them to specific phenotypes

What is Depmap?

Cancer Dependency Map

• By using functional genomic screens on thousands of cancer lines a map of cancer vulnerabilities could be created

• Assesses and measures all genes required for cell growth and drug sensitivity 

What is proteomic profiling and how is it useful for cancer research?

Comprehensive analysis of the entire set of proteins in a cell, tissue or organism

Can be used experimentally to identify biological functions 

• E.g. knockdown gene using shRNA then perform proteomic profiling

For cancer it is used to measure abundance of thousands of proteins in primary tumour samples 

• Used to discover new tumours or tumour subgroups 

• Used to define known tumours or tumour subgroups 

Why bother measuring proteins in cancer cells?

• Drug targets, signalling molecules and enzymes 

• Ultimately determine how a cancer will behave  

• Expression is subject to post-translational regulation

  • RNA alone cannot tell the whole story

How are AI and machine learning useful for cancer research?

• Handling complex omics data 

• Cancer subtyping and stratification  

• Predictive modelling 

• Integrating multi-modal data 

• Personalised medicine 

What is a supervised learning model?

Support vector machines and random forests predict clinical outcomes using labelled omics dataset 

• Key: Uses labelled data to predict outcomes

  • Known inputs and outputs

What is an unsupervised learning model?

Clustering algorithms like NMF, DBSCAN identify novel patient subgroups from unlabelled omics data 

• Key: Uses unlabelled data

  • Finds hidden patterns or structures on its own

What is a deep learning approach to cancer?

CNNs and RNNs integrate multi-omics data, modelling complex biological interactions and protein structures 

• CNN - Convolutional Neural Networks (CNNs)

• RNN - Recurrent Neural Networks (RNNs)

What is explainable AI (XAI)?

XAI techniques improve model transparency, enabling clinical trust through interpretable decision insights  

• Provides insight into the ‘why’ behind AI decisions

• Allows humans to comprehend and audit the outputs of machine learning algorithms