Genomics and variant scoring · BEGINNER LESSON
CADD
Combined Annotation-Dependent Depletion (CADD) is a machine-learning score for human genetic variants. It combines many genomic annotations to estimate how likely a variant is to be deleterious, meaning harmful to biological function or fitness.
What CADD is and what it measures
CADD stands for Combined Annotation-Dependent Depletion. It is a score used in human genetics to measure the relative likelihood that a variant will be deleterious, meaning that it is more likely to reduce biological function or fitness. It can score single nucleotide variants, which are changes in one DNA base, and short insertions or deletions, often called indels. The score is used to rank variants so researchers can focus on the most suspicious candidates first.
[1] CADD: predicting the deleteriousness of variants ... - PMC - NIH [3] A general framework for estimating the relative pathogenicity ...
How the score is built
CADD combines many kinds of genomic information, including sequence context, gene model annotations, evolutionary constraint, epigenetic measurements, and functional predictions. A machine-learning model is trained using two large groups of variants: a proxy-neutral group and a proxy-deleterious group. The proxy-neutral group contains human variants that became common or fixed after the human-chimpanzee split and are assumed to be mostly neutral or only weakly harmful. The proxy-deleterious group contains simulated variants that did not have to survive natural selection.
[1] CADD: predicting the deleteriousness of variants ... - PMC - NIH [3] A general framework for estimating the relative pathogenicity ...
How to read the output
After training, the model is applied to possible variants in the human reference genome and produces raw scores. Higher raw scores mean the variant looks more like the proxy-deleterious set than the proxy-neutral set. CADD also provides PHRED-scaled scores, which are normalized so results can be compared across the full set of possible single nucleotide variants. The main use is prioritizing variants among thousands to millions of candidates.
[1] CADD: predicting the deleteriousness of variants ... - PMC - NIH
Limitations and careful interpretation
CADD is useful, but its labels are not perfect. Some variants in the proxy-deleterious training set are actually neutral, so the training labels are only an approximation. CADD scores are also influenced by factors such as local mutation rate and background selection, and they do not precisely tell whether a variant is pathogenic, meaning disease-causing. The score is best treated as one line of evidence, not a final answer.
[1] CADD: predicting the deleteriousness of variants ... - PMC - NIH [3] A general framework for estimating the relative pathogenicity ...
Related idea in drug design
The same acronym CADD is also used in a different field to mean computer-aided drug design, which uses computers to help study drug-receptor interactions and predict binding. In this lesson, however, CADD means Combined Annotation-Dependent Depletion because that is the meaning supported by the genetic-variant sources.
[2] Computer Aided Drug Design and its Application to the Development of ...pmc.ncbi.nlm.nih.gov › articles › PMC6080097 [1] CADD: predicting the deleteriousness of variants ... - PMC - NIH
SEE THE IDEA
How CADD works
A human single nucleotide variant or short insertion/deletion to be scored
Sequence context, evolutionary, epigenetic, and functional features
Trained to separate proxy-neutral from proxy-deleterious variants
Raw or PHRED-scaled output used to rank likely deleterious variants
Input variants are described by many annotations, the model compares them with training sets, and the result is a score for prioritization.
CADD turns many genomic clues into a ranking score, but the score is still only evidence, not a direct diagnosis.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.