Bioinformatics final flash cards

What is machine learning?

Making predictions or decisions from data by finding patterns in data rather than using a closed‑form mathematical formula.

Essence of machine learning

An underlying pattern exists that we cannot describe with a closed form; we use data (preferably lots) to learn it.

ML notation: x, y, X, Y

x = feature vector; y = target value; X = matrix of feature vectors; Y = vector of target values.

Unknown target function f

The ideal (unknown) mapping from X to Y that we aim to approximate with data.

Training samples

Observed historical pairs (x_i, y_i) used to learn a model.

Hypothesis set H

A set of candidate functions {h1, h2, ...} from which the learning algorithm selects a final hypothesis.

Learning algorithm A

The procedure that uses training samples to select a hypothesis g ∈ H.

Final hypothesis g

The model chosen by the learning algorithm to approximate the unknown target function.

Supervised learning

Learning from labeled data; includes classification and regression.

Unsupervised learning

Learning from unlabeled data; includes clustering, association rules, and dimension reduction.

Classification task

Predicting a discrete label or class for an input (e.g., recurrent cancer vs no cancer).

Regression task

Predicting a continuous numerical value or relationship between variables.

Clustering

Grouping similar unlabeled data points into clusters (e.g., cancer subtypes).

Association rules

Finding co‑occurrence or co‑regulation patterns (e.g., coexpressed genes).

Dimension reduction

Reducing the number of features while preserving important structure (e.g., PCA).

Structured data

Data that fits into rows and columns (ideal for many ML algorithms).

Semi‑structured data

Data with partial structure such as JSON, HTML, or XML.

Unstructured data

Data without inherent tabular structure (images, free text, audio); often used in deep learning.

Data wrangling (munging)

Cleaning, correcting, and handling missing or inaccurate data before modeling.

Integration (data)

Combining multiple data sources into a single dataset for analysis.

Reduction (feature)

Consolidating or removing attributes to reduce dimensionality before modeling.

Aggregation (data)

Collapsing data to a common level (e.g., averages or sums) for analysis.

Exploratory Data Analysis (EDA)

Descriptive statistics, visualization, transforms, and tests used to understand data before modeling.

Normalization and standardization

Transformations to put features on comparable scales (e.g., z‑score, min‑max).

Training / Validation / Testing split

Common split: ~70% training, 10% validation, 20% testing; train fits model, validation tunes hyperparameters, test evaluates final performance.

Accuracy (definition)

Number of correct predictions divided by total predictions; depends on data quality and balance.

Gold standard data

High‑quality labeled data used as the reference for training and evaluation.

Accuracy paradox

High overall accuracy can be misleading when classes are imbalanced; a model can be accurate yet fail on the minority class.

Confusion matrix

Table summarizing true positives, false positives, true negatives, and false negatives (useful for classification evaluation).

Precision and recall

Precision = TP/(TP+FP); Recall = TP/(TP+FN); useful when class imbalance matters.

ROC curve and AUC

ROC plots true positive rate vs false positive rate; AUC measures overall separability of classes.

Decision trees

Classification/regression algorithm that splits data by feature thresholds to form a tree of decisions.

Random forest

Ensemble of decision trees that reduces variance by averaging many trees trained on bootstrapped samples.

Gradient boosting machines

Ensemble method that builds models sequentially to correct errors of prior models (e.g., XGBoost, LightGBM).

k‑nearest neighbors (k‑NN)

Predicts label based on the labels of the k closest training examples in feature space.

Logistic regression

Linear model for binary classification that outputs probabilities via the logistic function.

Naive Bayes

Probabilistic classifier assuming conditional independence of features given the class.

Support Vector Machine (SVM)

Finds a hyperplane that maximizes margin between classes; can use kernels for nonlinearity.

Neural networks

Models composed of layers of interconnected units (neurons) that learn hierarchical representations.

SVR (Support Vector Regression)

SVM variant for regression tasks.

K‑means clustering

Partitioning algorithm that assigns points to k clusters by minimizing within‑cluster variance.

Hierarchical clustering

Builds nested clusters by merging or splitting clusters based on distance metrics.

Apriori algorithm

Classic algorithm for mining frequent itemsets and association rules.

FP‑tree (frequent pattern tree)

Data structure for efficient mining of frequent patterns and association rules.

Principal Component Analysis (PCA)

Linear dimension reduction technique that projects data onto orthogonal components capturing maximum variance.

Linear Discriminant Analysis (LDA)

Dimension reduction and classification method that maximizes class separability in projected space.

t‑SNE

Nonlinear dimension reduction technique for visualizing high‑dimensional data in 2D/3D.

Autoencoder

Neural network that learns compressed representations (encoding) and reconstructs inputs (decoding).

Data quality importance

Models can only be as good as the data they are trained on; poor labels or bias degrade performance.

Overfitting

Model fits training data too closely and fails to generalize to new data; regularization and validation help prevent it.

Underfitting

Model is too simple to capture underlying patterns; increases bias and poor performance on train and test.

Cross‑validation

Method to estimate model performance by splitting data into multiple train/validation folds (e.g., k‑fold CV).

Feature engineering

Creating or transforming features to improve model performance (domain knowledge often helps).

Bias and variance tradeoff

Bias: error from wrong assumptions; Variance: error from sensitivity to training data; balance is key.

Regularization

Techniques (L1, L2) that penalize model complexity to reduce overfitting.

Hyperparameter tuning

Selecting model parameters (not learned during training) using validation data or CV (grid/random search).

Ensemble methods

Combine multiple models (bagging, boosting, stacking) to improve predictive performance.

Feature selection

Selecting a subset of relevant features to reduce dimensionality and improve model interpretability.

Imbalanced data strategies

Use resampling, class weighting, or specialized metrics when classes are imbalanced.

6 standard criteria of life

List: Cellular organization; Homeostasis; Metabolism; Response to stimuli; Reproduction; Adaptation and evolution.

Cellular organization

Organisms are composed of one or more highly organized cells.

Homeostasis

Organisms regulate and maintain internal conditions within narrow limits.

Metabolism

Organisms carry out complex chemical reactions to sustain life.

Response to stimuli

Organisms can detect and respond to changes in their environment.

Reproduction

Organisms produce new individuals of the same species (asexual or sexual).

Adaptation and evolution

Populations change over time through natural selection and genetic variation.

Prokaryotic vs Eukaryotic

Prokaryotes lack a nucleus (nucleoid instead); eukaryotes have a membrane‑bound nucleus.

Nucleoid

Region in prokaryotes containing the genome (not membrane‑bound).

Organelles

Membrane‑bound or specialized cellular compartments performing distinct functions.

Non‑organelle structures

Examples: cell membrane, cell wall, cytoplasm, flagella, pilus, ribosome.

Cell membrane

Phospholipid bilayer with embedded proteins, channels, and cholesterol that separates inside from outside.

Phospholipid bilayer

Two layers of phospholipids with hydrophilic heads and hydrophobic tails forming the membrane.

Integral vs peripheral proteins

Integral proteins span the membrane; peripheral proteins attach to the surface.

Cell wall

Flexible structural outer layer providing support and protection; composition varies by organism.

Cytoplasm

Liquid component inside the cell membrane containing organelles and molecules; ~80% water.

Flagellum

Hairlike appendage providing motility by whipping; structure varies across species.

Pilus

Short hairlike appendage used for adhesion, sensing, and cell recognition.

Ribosome

Complex of RNA and proteins that translates mRNA into polypeptides (protein synthesis).

Cilia

Threadlike projections that can move cells or sense the environment; similar to pilus in function.

Chloroplast

Membrane‑bound organelle in plants and algae that performs photosynthesis and contains chlorophyll.

Cytoskeleton

Network of protein filaments (actin, microtubules, intermediate filaments) providing structure and transport.

Rough Endoplasmic Reticulum (RER)

Membrane network studded with ribosomes; major site of protein synthesis and folding.

Smooth Endoplasmic Reticulum (SER)

Membrane network involved in lipid and small molecule synthesis; fewer ribosomes.

Golgi apparatus

Receives, sorts, modifies, and packages molecules for transport within or outside the cell.

Mitochondria

Membrane‑bound organelle that generates ATP; contains its own genome and is maternally inherited.

Vesicles

Small membrane‑bound compartments (endosomes, lysosomes, peroxisomes, vacuoles) for transport and storage.

Endosome vs lysosome

Endosomes sort internalized material; lysosomes contain degradative enzymes for breakdown.

Peroxisome

Organelle involved in oxidative reactions and detoxification (e.g., hydrogen peroxide metabolism).

Exosome

Small extracellular vesicle involved in intercellular communication and transport of molecules.

Macromolecule

definition: large polymeric biological molecules such as nucleic acids, proteins, carbohydrates, and lipids.

Monomer vs polymer

Monomer = single subunit (e.g., nucleotide, amino acid); Polymer = chain of monomers (e.g., DNA, protein).

Water (biological role)

Solvent for biological systems, polar molecule that stabilizes charged species and buffers temperature changes.

Carbohydrates

Composed of C, H, O; primary energy source; monomer = monosaccharide (e.g., glucose).

Glucose (formula)

Common sugar used for energy: C6H12O6.

Disaccharide example

Lactose is a disaccharide composed of two monosaccharides.

Polysaccharides

Glycogen (animal energy storage) and cellulose (plant structural component).

Lipids

Nonpolar molecules (fatty acids, phospholipids, sterols) used for energy storage, membranes, and signaling.

Fatty acid

structure: long hydrocarbon chain with a terminal carboxyl group; building block of many lipids.

Phospholipid

structural lipid with hydrophilic head and hydrophobic tails; forms cell membranes.

Sterols (cholesterol)

Rigid lipid molecules that modulate membrane fluidity and serve as hormone precursors.