eduKateSG · Why Science?
Move from molecular maps to mechanisms—connecting genetics, multi-omics, cohorts, functional validation, computation, clinical utility, ethics and equitable care
A fact-checked guide to the NUS Medicine Precision Medicine Translational Research Programme and its evidence chain from multi-omics discovery to decisions.
The official NUS programme uses multi-omics technologies to advance understanding of human-disease mechanisms and integrates molecular information with environmental, lifestyle and healthcare data.
Its current areas include human genetics, metabolic diseases and cohorts; functional genomics; metabolomics and lipidomics; computational biology and bioinformatics; and nucleic-acid technologies.
Did you know? A molecular feature can be associated with disease yet fail as a therapeutic target until experiments establish mechanism and modifiability.
This guide does not interpret personal genetic or medical data. NUS and qualified teams control current research and clinical implementation. Sources were checked 11 November 2026.
Chapter index
Chapters 1–9: precision medicine, cohorts, phenotype, specimens, genomics, functional genomics, transcriptomics, proteomics, metabolomics
Chapters 10–17: lipidomics, nucleic-acid technologies, bioinformatics, AI, biomarkers, utility, privacy, equity and evidence receipts
Individual variation can improve mechanism and decision evidence. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 1 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
State the decision that added information should improve. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
More detail does not automatically produce benefit. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Define precision medicine”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Write a question with a comparator. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 1 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Cohorts connect people, exposures, samples and outcomes. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 2 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Design recruitment and follow-up for the target population. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Convenient samples can encode bias. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Build representative cohorts”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Audit a recruitment plan. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 2 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Phenotypes include traits, diagnoses, physiology and outcomes. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 3 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Use consistent definitions and timing. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
A billing code may not equal biological state. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Measure phenotype carefully”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Create a phenotype dictionary. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 3 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Collection, processing, storage and freeze-thaw history affect measurements. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 4 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Use standard procedures and quality controls. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
An assay cannot recover a degraded sample. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Protect specimen integrity”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Draw a chain of custody. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 4 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
| Evidence move | What to record | What it cannot prove alone |
|---|---|---|
| Cohort question | Population, variable, unit and time window | A definition does not establish a result |
| Comparison | Baseline, comparator and uncertainty | Association alone does not establish cause |
| Decision boundary | Owner, trigger, review and escalation | A model does not replace accountable judgement |
Genomes contain common, rare, inherited and acquired variation. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 5 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Combine association with population and functional evidence. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
A variant is not deterministic. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Read genomic variation”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Classify fictional variants. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 5 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Perturbation tests how genes and regulatory regions affect function. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 6 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Use controls and rescue experiments. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
A cellular effect may not explain disease. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Use functional genomics”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Design a control set. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 6 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Transcriptomics describes RNA abundance across samples or cells. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 7 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Control batch, composition and multiple testing. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Expression difference does not prove causality. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Measure gene expression”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Compare bulk and single-cell questions. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 7 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Proteomics studies abundance, modification, localisation and interaction. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 8 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Validate detection and dynamic range. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
RNA level cannot always predict protein activity. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Measure proteins”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Build a validation ladder. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 8 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
| Evidence move | What to record | What it cannot prove alone |
|---|---|---|
| Omics question | Population, variable, unit and time window | A definition does not establish a result |
| Comparison | Baseline, comparator and uncertainty | Association alone does not establish cause |
| Decision boundary | Owner, trigger, review and escalation | A model does not replace accountable judgement |
Metabolomics and lipidomics capture products and intermediates. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 9 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Standardise collection and use reference materials. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
A peak can be associated before identity is secure. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Measure metabolites and lipids”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Trace feature identification. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 9 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
DNA and RNA tools can measure, perturb or potentially treat processes. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 10 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Test delivery, specificity, dose and safety. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Target engagement does not establish benefit. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Develop nucleic-acid technologies”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Draw an evidence gate. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 10 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Integration seeks relationships across molecular layers and context. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 11 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Predefine models and test added value. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Complexity can create unstable stories. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Integrate multi-omics”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Compare integration strategies. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 11 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Computational methods organise, predict and prioritise patterns. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 12 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Version code, references and environments. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Plausible output may reflect leakage. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Apply bioinformatics and AI”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Create a reproducibility checklist. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 12 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
| Evidence move | What to record | What it cannot prove alone |
|---|---|---|
| Mechanism question | Population, variable, unit and time window | A definition does not establish a result |
| Comparison | Baseline, comparator and uncertainty | Association alone does not establish cause |
| Decision boundary | Owner, trigger, review and escalation | A model does not replace accountable judgement |
Mechanism explains how change produces effect through a chain. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 13 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Use perturbation, rescue, temporal order and convergence. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Correlation maps nominate hypotheses. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Move from maps to mechanisms”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Design a mechanism triangle. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 13 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
A biomarker needs analytical validity, clinical validity and utility. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 14 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Set thresholds and evaluate calibration. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Significance does not justify a threshold. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Validate biomarkers”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Compare fictional distributions. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 14 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Utility asks whether information changes management and outcomes. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 15 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Compare the precision pathway with current practice. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
Better prediction may add cost or harm. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Evaluate clinical utility”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Build a net-benefit table. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 15 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
Molecular data can identify people and performance varies across groups. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 16 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Use governance, minimisation and representative validation. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
One ancestry category cannot capture all variation. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Protect privacy and equity”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Draft an audit. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 16 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
The receipt links cohort, sample, assay, computation, validation and decision. Precision medicine combines molecular information with environment, lifestyle and healthcare data to understand disease mechanisms and improve decisions. The NUS programme highlights human genetics, functional genomics, metabolomics, lipidomics, computational biology and nucleic-acid technologies. Chapter 17 asks what added information changes a real question, rather than assuming that collecting more measurements is automatically better.
Archive consent, provenance, versions and negative results. Reproducible work records consent, cohort selection, specimen handling, assay batch, reference build, feature definitions, quality thresholds, statistical model and code version. Thousands of variables may be tested, so researchers plan correction, replication and validation before interpreting a signal. They also compare the complex model with a simpler baseline and inspect calibration as well as discrimination.
A complete receipt does not turn probability into certainty. Precision does not mean certainty or a unique treatment for every person. A variant may have uncertain significance, a molecular pattern may reflect ancestry or environment, and a predictive association may not reveal a causal target. Clinical use requires validated assays, representative evidence, privacy protection, expert interpretation and proof that the information improves a meaningful outcome or decision.
For “Keep a precision medicine receipt”, build a synthetic table for ten imaginary people with a few harmless variables. Decide which question can be answered, reserve rows for checking and explain why a pattern might fail in another group. Assemble a one-page receipt. Never use real genetic or health data, and do not infer anyone’s ancestry, disease risk or treatment from a classroom activity.
Did you know? Adding features can make a model fit its development data better while making future predictions worse. Chapter 17 uses overfitting to connect molecular science with mathematics, computing and responsible evaluation. A valuable result must survive new samples, new assay batches and the ordinary messiness of clinical practice.
The capability map includes genetics, biochemistry, analytical chemistry, data science, bioinformatics, medicine, epidemiology, ethics and health systems. Admissions and professional roles remain governed by current official requirements. For learners, the core habit is tracing a claim from person and sample through measurement and analysis to a decision with a stated limit.
| Evidence move | What to record | What it cannot prove alone |
|---|---|---|
| Precision medicine question | Population, variable, unit and time window | A definition does not establish a result |
| Comparison | Baseline, comparator and uncertainty | Association alone does not establish cause |
| Decision boundary | Owner, trigger, review and escalation | A model does not replace accountable judgement |
Contents · Previous chapter · Continue in the Science Learning Hub
Primary sources and verification date
Source status for Why Science? | Precision Medicine Translational Research Programme was checked 11 November 2026. Research priorities, programmes, recruitment, facilities and participation routes can change; use the linked official owner for the current edition.
- NUS Precision Medicine TRP — Official vision, mission and current areas.
- Precision Medicine programme structure — Official objectives from maps to mechanisms.
- NUS Medicine research programmes — Official programme listing.
- NUS Medicine graduate research projects — Official current TRP research-project grouping.
Continue the evidence journey
Precision medicine becomes meaningful when molecular detail is tied to mechanisms, representative evidence, validated decisions and equitable outcomes rather than certainty theatre. Families considering science tuition or a science tutor can use the same evidence questions: what will the learner practise, what feedback will they receive, and how will progress be shown?