Bioinformatics: Al & Machine Learning . Dimensionality reduction refers to techniques that reduce the number of input variables(dimensionality/# of features) in a dataset. · Clustering involves automatically discovering natural grouping in data(commonalities) by interpreting only the input data. · Clustering is done to reveal the structure of the data(identify similar groups). . Dimensionality reduction is often motivated by computational concerns and complexity of ML(machine learning) models; it tells the model which features are redundant or carry less relevant information. · Goal of unsupervised learning: Discovering hidden patterns within a dataset; unsupervised learning may also have a goal of feature learning. · Feature learning allows the computational machine to automatically discover the representations that are needed to classify raw data. Supervised Learning · If there are continuous variables, use regression. · Classification = categorization Unsupervised Learning · Supervised learning has labels; unsupervised learning doesn't have labels. Computational Challenges in NGS Sequencing Data and Analysis · Alignment procedures are imperfect because a short DNA fragment is taken and mapped onto the human genome(a vector with 3 billion letters). Genotypes · Genotvne: Individual's collection of genes
-- . Each chromosome has two copies; each one coming from the mother and father. Normal Allelic Count Data . For allelic accounts, aN(or aT) = # of mismatches compared to the reference genome and dN(or dT) = # of reads. · When sequencing DNA, the origin can be unknown · A genotype with similar copies(AA or BB) has either zero mismatches or the same number of mismatches compared to reads(Ex. 6 mismatches / 6 reads). · A genotype with different copies(Ex. AB) can have approximately half of the mismatches compared to reads(4/7 or 3/6) · If a base of a reference genome is different from the base of the normal genome, there is a mismatch. Binomial Distribution for Normal Allelic Count Data · K-number of observations that are mismatches · P-chance of mismatches · N=number of trials/experiment trials · P=probability of a mismatch Normal-Tumour Allelic Count Data . If the number of mismatches increases from a normal sample to a tumor sample, it's most likely a cancer mutation that is also a germline mutation. . When the number of mismatches decreases from a normal sample to a tumor sample, it's most likely a somatic mutation. ROC Curves(Receiver Operating Characteristic) · Higher sensitivity is better. TP53:
· The most commonly mutated driver gene in cancer. Extra Vocab: Confer-grant or bestow Metastatic-Referring to metastasis: a pathogenic agent's spread from an initial site to a secondary site within the host's body Stochastic: randomly determined; may not be predicted precisely Neoplastic: Related to neoplasm-an abnormal growth of cells Chemotherapeutic: Referring to a type of cancer treatment used to treat or cure cancers. NGS: · The interpretable information in NGS-based bioinformatics are genomic variants. · If there is no reference genome, use assembly.