Artificial Intelligence Foundations for Biomedical Research
Expert-defined terms from the Professional Certificate in Artificial Intelligence in Biomedical Research course at LearnUNI. Free to read, free to share, paired with a professional course.
Algorithm A step‑by‑step computational procedure for solving a problem or… #
Algorithm A step‑by‑step computational procedure for solving a problem or performing a task.
Example #
A decision‑tree algorithm classifies patient disease status based on clinical variables.
Application #
Predicting drug response from genomic data.
Challenge #
Choosing an algorithm that balances accuracy with interpretability for clinicians.
Artificial Neural Network (ANN) A computational model inspired by the bra… #
Artificial Neural Network (ANN) A computational model inspired by the brain’s network of neurons, consisting of layers of interconnected nodes that learn patterns from data.
Example #
A multilayer perceptron predicts protein secondary structure from amino‑acid sequences.
Application #
Modeling complex genotype‑phenotype relationships.
Challenge #
Requires large labeled datasets and can be a “black box” for regulatory approval.
Artificial Intelligence (AI) The broader field encompassing machines that… #
Artificial Intelligence (AI) The broader field encompassing machines that can perform tasks requiring human intelligence, such as reasoning, learning, and perception.
Example #
AI‑driven chatbots triage patient inquiries in telemedicine platforms.
Application #
Accelerating drug discovery through virtual screening.
Challenge #
Ethical considerations, bias mitigation, and ensuring fairness across diverse populations.
Association Rule Mining A data‑mining technique that discovers interestin… #
Association Rule Mining A data‑mining technique that discovers interesting relationships (rules) between variables in large datasets.
Example #
Identifying that patients with hypertension and high cholesterol often develop cardiovascular disease.
Application #
Informing preventive care pathways.
Challenge #
Generating clinically relevant rules without overwhelming clinicians with spurious associations.
Autoencoder An unsupervised neural network that learns to compress data i… #
Autoencoder An unsupervised neural network that learns to compress data into a lower‑dimensional representation and then reconstruct it.
Example #
Reducing noise in single‑cell RNA‑seq data while preserving biological signal.
Application #
Feature extraction for downstream classification tasks.
Challenge #
Preventing over‑compression that removes biologically important variation.
Batch Effect Systematic non‑biological variation introduced during sample… #
Batch Effect Systematic non‑biological variation introduced during sample processing, measurement, or data acquisition.
Example #
Differences in gene‑expression levels caused by using two different sequencing platforms.
Application #
Correcting batch effects improves reproducibility of biomarker studies.
Challenge #
Detecting subtle batch effects without removing true biological differences.
Bayesian Inference A statistical method that updates the probability of a… #
Bayesian Inference A statistical method that updates the probability of a hypothesis as more evidence becomes available, using prior distributions and likelihood functions.
Example #
Estimating the probability that a variant is pathogenic given population frequency and functional assay results.
Application #
Personalized risk prediction models.
Challenge #
Choosing appropriate priors and computationally scaling to high‑dimensional biomedical data.
Bias (Algorithmic) Systematic error that leads to unfair outcomes for cer… #
Bias (Algorithmic) Systematic error that leads to unfair outcomes for certain groups, often arising from biased training data or model design.
Example #
A predictive model underestimates disease risk for under‑represented ethnic minorities.
Application #
Adjusting models to ensure equitable healthcare delivery.
Challenge #
Detecting hidden biases and implementing corrective strategies without sacrificing performance.
Biological Ontology Structured vocabularies that describe entities and re… #
Biological Ontology Structured vocabularies that describe entities and relationships in biology, enabling standardized data annotation.
Example #
Using the Disease Ontology to map clinical diagnoses to standardized terms for machine‑learning pipelines.
Application #
Facilitating data integration across heterogeneous biomedical databases.
Challenge #
Maintaining up‑to‑date ontologies and handling ambiguous annotations.
Biomarker A measurable indicator of a biological state, disease progressi… #
Biomarker A measurable indicator of a biological state, disease progression, or therapeutic response.
Example #
Circulating tumor DNA levels as a biomarker for treatment efficacy in oncology.
Application #
Guiding precision‑medicine decisions.
Challenge #
Validating biomarkers across cohorts and ensuring clinical utility.
Black‑Box Model An AI model whose internal decision logic is not readily… #
Black‑Box Model An AI model whose internal decision logic is not readily interpretable by humans.
Example #
A deep convolutional network predicting pathology from imaging data without clear feature attribution.
Application #
High‑accuracy diagnostic tools.
Challenge #
Regulatory bodies often require transparent reasoning for clinical adoption.
Bootstrap Aggregating (Bagging) An ensemble technique that builds multipl… #
Bootstrap Aggregating (Bagging) An ensemble technique that builds multiple models on random subsets of data and aggregates their predictions to improve stability.
Example #
Bagging decision trees to predict patient survival from electronic health records.
Application #
Reducing over‑fitting in high‑dimensional biomedical datasets.
Challenge #
Increased computational cost and difficulty in interpreting individual model contributions.
Calibration (Model) The agreement between predicted probabilities and obs… #
Calibration (Model) The agreement between predicted probabilities and observed outcomes; a well‑calibrated model provides reliable risk estimates.
Example #
Adjusting a logistic‑regression model so that a predicted 10% risk of disease matches actual incidence.
Application #
Clinical decision support where absolute risk informs treatment thresholds.
Challenge #
Maintaining calibration when models are transferred to new populations.
Clustering An unsupervised learning technique that groups similar observa… #
Clustering An unsupervised learning technique that groups similar observations together based on feature similarity.
Example #
Grouping patients by gene‑expression profiles to identify molecular subtypes of breast cancer.
Application #
Discovering disease phenotypes for targeted therapies.
Challenge #
Determining the appropriate number of clusters and validating biological relevance.
Convolutional Neural Network (CNN) A deep learning architecture specializ… #
Convolutional Neural Network (CNN) A deep learning architecture specialized for processing grid‑like data such as images, using convolutional filters to capture spatial hierarchies.
Example #
A CNN classifies histopathology slides into cancer grades.
Application #
Automated image‑based diagnostics.
Challenge #
Requires large annotated image datasets and careful handling of domain shift between institutions.
Cross‑Validation A resampling method for assessing model performance by d… #
Cross‑Validation A resampling method for assessing model performance by dividing data into training and testing folds multiple times.
Example #
5‑fold cross‑validation evaluates a classifier predicting adverse drug reactions.
Application #
Preventing over‑optimistic performance estimates in limited biomedical cohorts.
Challenge #
Computational overhead for complex models and ensuring folds respect patient independence.
Data Augmentation Techniques that artificially expand training datasets b… #
Data Augmentation Techniques that artificially expand training datasets by applying transformations to existing samples.
Example #
Rotating and flipping MRI scans to improve robustness of a segmentation model.
Application #
Enhancing model generalizability when limited patient images are available.
Challenge #
Augmentations must preserve underlying biological meaning to avoid misleading the model.
Data Imbalance Situations where certain classes or outcomes are under‑rep… #
Data Imbalance Situations where certain classes or outcomes are under‑represented relative to others, common in rare‑disease datasets.
Example #
Fewer positive cases of a rare genetic disorder compared to negative controls in a training set.
Application #
Adjusting model training to detect rare events.
Challenge #
Preventing over‑fitting to minority class while maintaining overall accuracy.
Data Integration The process of combining heterogeneous biomedical datase… #
g., genomics, imaging, clinical records) into a unified analytical framework.
Example #
Merging transcriptomic and proteomic profiles to improve cancer subtype prediction.
Application #
Holistic patient phenotyping for precision medicine.
Challenge #
Aligning disparate data formats, handling missingness, and preserving privacy.
Deep Learning A subset of machine learning that uses neural networks with… #
Deep Learning A subset of machine learning that uses neural networks with many layers to automatically learn hierarchical feature representations.
Example #
A deep recurrent network predicts disease trajectories from longitudinal health records.
Application #
Modeling complex temporal patterns in patient data.
Challenge #
High computational demand, need for large labeled datasets, and interpretability concerns.
Dimensionality Reduction Techniques that reduce the number of variables w… #
Dimensionality Reduction Techniques that reduce the number of variables while retaining essential information, facilitating visualization and downstream modeling.
Example #
Principal component analysis compresses gene‑expression matrices for clustering analysis.
Application #
Simplifying high‑throughput omics data for pattern discovery.
Challenge #
Balancing information loss against computational tractability and biological interpretability.
Discriminative Model A model that learns the decision boundary between cl… #
Discriminative Model A model that learns the decision boundary between classes directly, focusing on predicting the target variable.
Example #
A support vector machine separates patients with and without sepsis based on vital signs.
Application #
Rapid risk stratification in intensive‑care units.
Challenge #
May require careful feature engineering to capture complex interactions.
Domain Adaptation Methods that adjust models trained on one data distribu… #
Domain Adaptation Methods that adjust models trained on one data distribution (source domain) to perform well on a different but related distribution (target domain).
Example #
Fine‑tuning a CNN trained on public chest‑X‑ray datasets to work on images from a local hospital.
Application #
Reducing the need for extensive annotation in each new clinical setting.
Challenge #
Detecting and correcting subtle distributional differences without degrading performance.
Ensemble Learning Combining multiple models to produce a single, often mo… #
Ensemble Learning Combining multiple models to produce a single, often more robust, prediction.
Example #
An ensemble of gradient‑boosted trees and neural networks predicts patient readmission risk.
Application #
Improving predictive accuracy for complex clinical outcomes.
Challenge #
Managing increased model complexity and computational resources.
Feature Engineering The process of selecting, transforming, and construct… #
Feature Engineering The process of selecting, transforming, and constructing variables that improve model performance.
Example #
Converting medication codes into binary indicators and calculating dosage‑frequency metrics.
Application #
Enhancing model interpretability and reducing over‑fitting.
Challenge #
Labor‑intensive and may require domain expertise to capture relevant biological signals.
Feature Selection Techniques that identify a subset of informative variab… #
Feature Selection Techniques that identify a subset of informative variables from a larger set, reducing dimensionality and improving model efficiency.
Example #
Using mutual information to select the top 50 SNPs associated with a disease phenotype.
Application #
Streamlining predictive models for clinical deployment.
Challenge #
Avoiding loss of predictive power while eliminating redundant or noisy features.
Gene Ontology (GO) A hierarchical vocabulary describing gene product attr… #
Gene Ontology (GO) A hierarchical vocabulary describing gene product attributes across three domains: biological process, molecular function, and cellular component.
Example #
Enriching a list of differentially expressed genes with GO terms reveals involvement in immune response.
Application #
Interpreting omics results in a biologically meaningful context.
Challenge #
Managing multiple testing corrections and varying annotation quality across species.
Gradient Descent An optimization algorithm that iteratively adjusts model… #
Gradient Descent An optimization algorithm that iteratively adjusts model parameters to minimize a loss function by moving in the direction of steepest descent.
Example #
Training a neural network to predict drug toxicity by minimizing mean‑squared error.
Application #
Core method for fitting many AI models in biomedical research.
Challenge #
Selecting appropriate step sizes and avoiding local minima in non‑convex landscapes.
Graph Neural Network (GNN) Neural architectures designed to operate on gr… #
Graph Neural Network (GNN) Neural architectures designed to operate on graph‑structured data, propagating information across nodes and edges.
Example #
Predicting protein‑protein interaction strength using a GNN that incorporates known interaction networks.
Application #
Modeling biological pathways and molecular interaction maps.
Challenge #
Scaling to large, sparse biomedical graphs and interpreting learned representations.
Health Informatics The interdisciplinary field that uses information tech… #
Health Informatics The interdisciplinary field that uses information technology to acquire, store, retrieve, and use health data for improved patient care.
Example #
Integrating AI‑driven alerts into an EHR to flag potential drug‑drug interactions.
Application #
Enhancing safety and efficiency in clinical workflows.
Challenge #
Ensuring interoperability, data privacy, and clinician acceptance.
Imbalanced Learning Strategies specifically designed to address class imb… #
Imbalanced Learning Strategies specifically designed to address class imbalance, such as resampling, cost‑sensitive learning, and synthetic minority oversampling.
Example #
Applying SMOTE to generate synthetic cases of a rare adverse event before training a classifier.
Application #
Improving detection of low‑prevalence conditions.
Challenge #
Synthetic samples must reflect realistic biological variation to avoid misleading the model.
Inference (Statistical) The process of drawing conclusions about a popula… #
Inference (Statistical) The process of drawing conclusions about a population based on sample data, often involving hypothesis testing or estimation.
Example #
Inferring that a new biomarker is significantly associated with survival after correcting for covariates.
Application #
Validating AI‑derived hypotheses in biomedical research.
Challenge #
Controlling false discovery rates when testing thousands of genomic features.
Instance‑Based Learning Approaches that store training instances and make… #
Instance‑Based Learning Approaches that store training instances and make predictions by comparing new inputs to stored examples, rather than learning explicit parameters.
Example #
Using k‑NN to classify patient phenotypes based on similarity of laboratory values.
Application #
Simple, interpretable models for small clinical datasets.
Challenge #
Computationally intensive for large biomedical databases and sensitive to noisy measurements.
Integration (Multi‑omics) Combining data from multiple omics layers #
genomics, transcriptomics, proteomics, metabolomics—to gain a comprehensive view of biological systems.
Example #
Joint analysis of DNA‑methylation and RNA‑seq data identifies epigenetically regulated gene signatures in cancer.
Application #
More accurate disease subtyping and therapeutic target discovery.
Challenge #
Aligning disparate data scales, handling missing modalities, and ensuring reproducibility.
Interoperability The ability of different information systems, devices, o… #
Interoperability The ability of different information systems, devices, or applications to exchange and use data cohesively.
Example #
A machine‑learning pipeline that reads patient data via the HL7 FHIR API from diverse hospital EHRs.
Application #
Facilitating cross‑institutional AI studies and clinical deployment.
Challenge #
Varying data quality, inconsistent coding practices, and regulatory constraints.
Interpretability The degree to which a human can understand the internal… #
Interpretability The degree to which a human can understand the internal mechanics or decisions of an AI model.
Example #
SHAP values highlight that elevated C‑reactive protein contributes most to a sepsis risk score.
Application #
Building clinician trust and meeting regulatory requirements.
Challenge #
Balancing interpretability with predictive performance, especially for deep models.
k‑Nearest Neighbors (k‑NN) An instance‑based algorithm that classifies a… #
k‑Nearest Neighbors (k‑NN) An instance‑based algorithm that classifies a new sample based on the majority label among its k closest training examples.
Example #
Classifying tumor type using similarity of gene‑expression profiles to known reference samples.
Application #
Rapid prototyping of diagnostic classifiers.
Challenge #
Sensitivity to feature scaling and computational cost for large biomedical datasets.
Knowledge Graph A network‑based representation of entities (e #
g., genes, diseases, drugs) and their relationships, enabling semantic queries and reasoning.
Example #
A biomedical knowledge graph links a drug to its target protein, associated pathways, and observed side effects.
Application #
Drug repurposing by traversing connections between diseases and existing therapeutics.
Challenge #
Keeping the graph up‑to‑date and handling contradictory information from heterogeneous sources.
Learning Rate A hyperparameter that controls the step size during gradien… #
Learning Rate A hyperparameter that controls the step size during gradient‑descent optimization; too large can cause divergence, too small slows convergence.
Example #
Setting an initial learning rate of 0.001 for training a convolutional network on histology images.
Application #
Tuning model training to achieve stable and efficient convergence.
Challenge #
Selecting schedules (e.g., decay, warm‑up) that adapt to the complexity of biomedical data.
Logistic Regression A statistical model that predicts the probability of… #
Logistic Regression A statistical model that predicts the probability of a binary outcome using a linear combination of input features and a logistic link function.
Example #
Estimating the risk of myocardial infarction based on age, cholesterol level, and smoking status.
Application #
Baseline model for risk stratification in clinical studies.
Challenge #
Limited ability to capture non‑linear interactions without feature engineering.
Long Short‑Term Memory (LSTM) A recurrent neural network architecture tha… #
Long Short‑Term Memory (LSTM) A recurrent neural network architecture that mitigates vanishing gradients, enabling learning of long‑range dependencies in sequential data.
Example #
Predicting future lab values from a patient’s longitudinal health record using an LSTM.
Application #
Forecasting disease progression and treatment response.
Challenge #
Requires substantial sequential data and careful handling of irregular time intervals common in clinical settings.
Machine Learning (ML) A subset of AI that enables computers to learn patt… #
Machine Learning (ML) A subset of AI that enables computers to learn patterns from data without explicit programming, encompassing supervised, unsupervised, and reinforcement approaches.
Example #
Training a random forest to classify MRI scans as benign or malignant.
Application #
Automating diagnostic image analysis and risk prediction.
Challenge #
Ensuring generalizability across diverse patient populations and data sources.
Meta‑Learning “Learning to learn” approaches that aim to improve model ad… #
Meta‑Learning “Learning to learn” approaches that aim to improve model adaptability to new tasks with limited data, often through few‑shot or transfer techniques.
Example #
A meta‑learner quickly fine‑tunes a model to predict rare disease phenotypes using only a handful of cases.
Application #
Accelerating AI development for under‑studied conditions.
Challenge #
Designing meta‑training regimes that capture the breadth of biomedical variability.
Missing Data Imputation Strategies to estimate unobserved values in datas… #
Missing Data Imputation Strategies to estimate unobserved values in datasets, preserving analytical integrity.
Example #
Using chained equations to impute missing blood‑pressure readings in a cohort study.
Application #
Enabling complete‑case analysis for predictive modeling.
Challenge #
Imputation methods must respect the underlying biological mechanisms to avoid bias.
Multicollinearity A situation where two or more predictor variables are h… #
Multicollinearity A situation where two or more predictor variables are highly correlated, potentially destabilizing coefficient estimates in regression models.
Example #
Age and frailty scores are strongly correlated in a geriatric risk model.
Application #
Identifying and removing redundant features to improve model robustness.
Challenge #
Detecting subtle collinearity in high‑dimensional omics data and deciding on appropriate remedial actions.
Natural Language Processing (NLP) Techniques for analyzing, understanding… #
Natural Language Processing (NLP) Techniques for analyzing, understanding, and generating human language, applied to clinical notes, literature, and patient‑generated data.
Example #
Using a BERT‑based model to extract medication names and dosages from discharge summaries.
Application #
Automating phenotyping and adverse‑event detection from unstructured text.
Challenge #
Dealing with domain‑specific jargon, abbreviations, and privacy‑preserving de‑identification.
Neural Architecture Search (NAS) Automated methods for discovering optima… #
Neural Architecture Search (NAS) Automated methods for discovering optimal network structures tailored to a specific dataset or task.
Example #
NAS identifies a lightweight CNN architecture that achieves high accuracy on low‑resolution retinal images.
Application #
Customizing AI models for resource‑constrained clinical environments.
Challenge #
Computationally intensive and may produce architectures difficult to interpret biologically.
Normalization (Data) Transformations that adjust data to a common scale,… #
Normalization (Data) Transformations that adjust data to a common scale, facilitating comparison and model training.
Example #
Log‑transforming RNA‑seq counts before downstream analysis.
Application #
Reducing variance caused by differing measurement units across features.
Challenge #
Selecting appropriate normalization methods that preserve biological signal.
Object Detection Computer‑vision tasks that locate and classify multiple… #
Object Detection Computer‑vision tasks that locate and classify multiple objects within an image, often using bounding boxes or segmentation masks.
Example #
Detecting pulmonary nodules in chest CT scans for early‑lung‑cancer screening.
Application #
Assisting radiologists with automated lesion identification.
Challenge #
Achieving high sensitivity while maintaining low false‑positive rates in heterogeneous imaging datasets.
One‑Hot Encoding A representation method that converts categorical variab… #
One‑Hot Encoding A representation method that converts categorical variables into binary vectors, with each category assigned a unique position.
Example #
Encoding blood‑type (A, B, AB, O) as four binary columns for a predictive model.
Application #
Preparing clinical categorical data for machine‑learning algorithms that require numeric input.
Challenge #
High dimensionality when categories are numerous, potentially leading to sparse data.
Overfitting The phenomenon where a model captures noise or idiosyncrasies… #
Overfitting The phenomenon where a model captures noise or idiosyncrasies of the training data, resulting in poor performance on unseen data.
Example #
A deep network achieves 99% accuracy on training images but only 70% on external test sets.
Application #
Recognizing the need for robust validation in biomedical AI research.
Challenge #
Mitigating overfitting through techniques such as dropout, early stopping, and data augmentation.
Precision Medicine Tailoring medical treatment to the individual characte… #
Precision Medicine Tailoring medical treatment to the individual characteristics of each patient, often leveraging genomic, phenotypic, and lifestyle data.
Example #
Selecting an EGFR inhibitor for lung‑cancer patients with a specific mutation.
Application #
Improving therapeutic efficacy and reducing adverse effects.
Challenge #
Integrating heterogeneous data sources and ensuring equitable access to precision approaches.
Principal Component Analysis (PCA) A linear dimensionality‑reduction tech… #
Principal Component Analysis (PCA) A linear dimensionality‑reduction technique that transforms correlated variables into a set of uncorrelated components ordered by explained variance.
Example #
Reducing thousands of gene‑expression features to the top 10 principal components for clustering.
Application #
Visualizing high‑dimensional biomedical data and removing redundancy.
Challenge #
Linear nature may miss non‑linear biological relationships; interpretation of components can be non‑trivial.
Probabilistic Graphical Model A framework that represents conditional dep… #
Probabilistic Graphical Model A framework that represents conditional dependencies among random variables using graphs, enabling joint probability inference.
Example #
Modeling the probabilistic relationships between genetic variants, gene expression, and disease phenotype.
Application #
Integrating multi‑layered biological data with uncertainty quantification.
Challenge #
Learning structure and parameters from limited biomedical data can be computationally demanding.
Protein‑Protein Interaction (PPI) Network A graph representation of physi… #
Protein‑Protein Interaction (PPI) Network A graph representation of physical or functional interactions among proteins, often used to infer functional modules.
Example #
Analyzing a PPI network to identify central proteins that may serve as drug targets in cancer.
Application #
Systems‑level understanding of disease mechanisms.
Challenge #
Incomplete or noisy interaction data and the need for experimental validation.
Quantile Normalization A technique that forces the distribution of each s… #
Quantile Normalization A technique that forces the distribution of each sample’s measurements to be identical, commonly used in microarray and RNA‑seq analyses.
Example #
Aligning expression profiles across multiple tissue samples before differential expression testing.
Application #
Reducing technical variability in high‑throughput assays.
Challenge #
May obscure true biological differences if the underlying assumption of shared distribution is violated.
Random Forest An ensemble of decision trees where each tree is trained on… #
Random Forest An ensemble of decision trees where each tree is trained on a random subset of features and samples, and predictions are aggregated by majority vote or averaging.
Example #
Predicting patient mortality using a random forest that incorporates demographics, lab results, and comorbidities.
Application #
Robust, non‑linear modeling with built‑in variable importance metrics.
Challenge #
Large forests can be memory‑intensive and less interpretable than single trees.
Recurrent Neural Network (RNN) A class of neural networks designed for se… #
Recurrent Neural Network (RNN) A class of neural networks designed for sequential data, where connections form directed cycles allowing information persistence across time steps.
Example #
Modeling the progression of vital‑sign trajectories to forecast septic shock onset.
Application #
Analyzing longitudinal health records and time‑series biomarker data.
Challenge #
Training instability due to vanishing or exploding gradients; mitigated by architectures such as LSTM or GRU.
Regularization Techniques that add a penalty to the loss function to disc… #
Regularization Techniques that add a penalty to the loss function to discourage complex models, thereby reducing overfitting.
Example #
Applying L1 regularization (lasso) to shrink coefficients of irrelevant genes in a regression model.
Application #
Producing sparse, interpretable models for biomarker discovery.
Challenge #
Selecting appropriate regularization strength; overly strong penalties may eliminate true signals.
Reinforcement Learning (RL) An AI paradigm where an agent learns to make… #
Reinforcement Learning (RL) An AI paradigm where an agent learns to make sequential decisions by receiving rewards or penalties from its environment.
Example #
An RL agent optimizes chemotherapy dosing schedules to maximize tumor reduction while minimizing toxicity.
Application #
Adaptive treatment planning and dynamic clinical pathway optimization.
Challenge #
Defining realistic reward functions and ensuring safety in medical contexts.
Resampling Methods that repeatedly draw samples from a dataset to assess… #
Resampling Methods that repeatedly draw samples from a dataset to assess model stability, estimate variance, or address class imbalance.
Example #
Bootstrapping confidence intervals for the AUC of a diagnostic classifier.
Application #
Providing robust performance metrics for small biomedical cohorts.
Challenge #
Computational cost for deep models and maintaining independence between resampled subsets.
Risk Score A numeric value derived from predictive models that quantifies… #
Risk Score A numeric value derived from predictive models that quantifies an individual’s probability of a clinical event.
Example #
The CHA₂DS₂‑VASc score estimates stroke risk in atrial‑fibrillation patients.
Application #
Guiding therapeutic decisions such as anticoagulation initiation.
Challenge #
Translating complex AI outputs into intuitive, actionable scores for clinicians.
Scalable Computing Architectures and algorithms designed to handle increa… #
Scalable Computing Architectures and algorithms designed to handle increasing data volumes and computational demands without performance degradation.
Example #
Running a Spark‑based pipeline to process petabytes of whole‑genome sequencing data.
Application #
Enabling large‑scale biomedical AI projects across consortia.
Challenge #
Managing data security, cost, and reproducibility across heterogeneous compute environments.
Semantic Similarity Quantitative measures that assess the closeness of co… #
Semantic Similarity Quantitative measures that assess the closeness of concepts within an ontology or knowledge graph, often based on shared ancestors.
Example #
Computing similarity between “myocardial infarction” and “coronary artery disease” to cluster patient diagnoses.
Application #
Enhancing phenotype‑genotype association analyses.
Challenge #
Choosing appropriate similarity metrics and handling incomplete annotations.
Sequence Alignment The process of arranging DNA, RNA, or protein sequence… #
Sequence Alignment The process of arranging DNA, RNA, or protein sequences to identify regions of similarity that may indicate functional, structural, or evolutionary relationships.
Example #
Aligning viral genomes to detect conserved regions for vaccine target design.
Application #
Comparative genomics and mutation impact prediction.
Challenge #
Computational intensity for large‑scale datasets and handling indels in highly variable regions.
Signal‑to‑Noise Ratio (SNR) A measure comparing the level of a desired si… #
Signal‑to‑Noise Ratio (SNR) A measure comparing the level of a desired signal to the level of background noise, critical for evaluating assay quality.
Example #
A high SNR in mass‑spectrometry data indicates reliable detection of low‑abundance metabolites.
Application #
Assessing data quality before model training.
Challenge #
Improving SNR in noisy clinical measurements without sacrificing patient comfort.
Single‑Cell RNA‑seq (scRNA‑seq) High‑throughput technology that profiles… #
Single‑Cell RNA‑seq (scRNA‑seq) High‑throughput technology that profiles gene expression at the resolution of individual cells, revealing cellular heterogeneity.
Example #
Identifying tumor‑infiltrating immune cell subpopulations that predict response to immunotherapy.
Application #
Discovering novel cell types and therapeutic targets.
Challenge #
Managing dropout events, batch effects, and computationally intensive clustering.
Smoothing (Time Series) Techniques that reduce short‑term fluctuations to… #
Smoothing (Time Series) Techniques that reduce short‑term fluctuations to reveal underlying trends in sequential biomedical measurements.
Example #
Applying a 7‑day moving average to daily blood‑glucose readings to visualize long‑term control.
Application #
Enhancing signal clarity for predictive modeling of disease trajectories.
Challenge #
Selecting window sizes that preserve clinically relevant variability.
Softmax Function A mathematical transformation that converts a vector of… #
Softmax Function A mathematical transformation that converts a vector of raw scores into probabilities that sum to one, commonly used in multi‑class classification.
Example #
Output layer of a CNN assigns probabilities to categories of tissue pathology.
Application #
Providing interpretable confidence scores for diagnostic AI systems.
Challenge #
Calibration of softmax outputs to reflect true likelihoods in clinical settings.
Sparse Coding Representing data as a linear combination of a few basis el… #
Sparse Coding Representing data as a linear combination of a few basis elements, promoting parsimonious representations.
Example #
Encoding MRI patches using a sparse dictionary to improve reconstruction speed.
Application #
Efficient storage and transmission of high‑resolution medical images.
Challenge #
Learning dictionaries that capture relevant anatomical structures without excessive computational burden.
Supervised Learning Machine‑learning tasks that use labeled input–output… #
Supervised Learning Machine‑learning tasks that use labeled input–output pairs to learn a mapping from features to targets.
Example #
Training a model to predict cancer stage from imaging biomarkers labeled by radiologists.
Application #
Automated diagnostic decision support.
Challenge #
Obtaining high‑quality labeled data, especially for rare diseases.
Support Vector Machine (SVM) A discriminative classifier that finds the h… #
Support Vector Machine (SVM) A discriminative classifier that finds the hyperplane maximizing margin between classes, often using kernel functions to handle non‑linear separation.
Example #
An SVM with a radial basis function kernel distinguishes malignant from benign breast lesions based on ultrasound features.
Application #
Robust classification with limited training samples.
Challenge #
Scaling to large biomedical datasets and selecting appropriate kernel parameters.
Survival Analysis Statistical methods for analyzing time‑to‑event data, a… #
Survival Analysis Statistical methods for analyzing time‑to‑event data, accounting for censored observations.
Example #
Modeling the effect of a new therapy on overall survival of metastatic cancer patients.
Application #
Evaluating treatment efficacy and risk stratification.
Challenge #
Incorporating high‑dimensional genomic covariates while satisfying model assumptions.
Transfer Learning Leveraging knowledge from a source task (often with abu… #
Transfer Learning Leveraging knowledge from a source task (often with abundant data) to improve learning on a target task with limited data.
Example #
Adapting a CNN pre‑trained on ImageNet to classify histopathology slides after retraining the final layers.
Application #
Reducing annotation burden in specialized biomedical imaging domains.
Challenge #
Domain mismatch can lead to negative transfer if source and target data differ substantially.
t‑Distributed Stochastic Neighbor Embedding (t‑SNE) A non‑linear dimensio… #
t‑Distributed Stochastic Neighbor Embedding (t‑SNE) A non‑linear dimensionality‑reduction technique that visualizes high‑dimensional data by preserving local structure in a low‑dimensional space.
Example #
Plotting single‑cell transcriptomic profiles to reveal distinct cell clusters.
Application #
Exploratory data analysis for hypothesis generation.
Challenge #
Computationally intensive for large datasets and sensitive to hyper‑parameter choices.
Temporal Fusion Transformer (TFT) An advanced deep‑learning architecture… #
Temporal Fusion Transformer (TFT) An advanced deep‑learning architecture that integrates static covariates, known future inputs, and past observations for multivariate time‑series forecasting.
Example #
Predicting future laboratory values for intensive‑care patients using TFT to incorporate medication schedules and vital signs.
Application #
Dynamic risk prediction and personalized monitoring.
Challenge #
Requires extensive temporal data and careful handling of missing values.
Uncertainty Quantification Techniques that estimate the confidence or var… #
Uncertainty Quantification Techniques that estimate the confidence or variability of model predictions, crucial for risk‑aware clinical decision making.
Example #
Providing predictive intervals for disease progression models to inform treatment planning.
Application #
Enhancing trust and safety in AI‑assisted healthcare.
Challenge #
Balancing computational cost with the fidelity of uncertainty estimates.
Unsupervised Learning Methods that infer structure from unlabeled data, d… #
Unsupervised Learning Methods that infer structure from unlabeled data, discovering hidden patterns, clusters, or representations.
Example #
Example