Loading animations
Some things are far easier to watch than to read about. These explainers cover the concepts students get stuck on — how sequencing actually works, why we run FastQC, how a read finds its place in the genome, and how a neural network learns.
Algorithms
Watch a dynamic-programming matrix fill one cell at a time, then trace back to read the alignment out. Change the gap penalty and see a different alignment win.
Algorithms
Shred reads into k-mers, build the graph they imply, and walk it to rebuild the genome. Move the k slider and watch the graph tangle or fragment.
Algorithms
The same two sequences, the same scores, two matrices side by side. Only two lines of the algorithm differ — and they turn out to be the difference between a sequence comparison and a database search.
Algorithms
Five gap characters, arranged two ways. A linear penalty scores them identically; an affine one does not. Watch the gap-open slider decide which alignment your tools report.
Algorithms
Rotate a string every way, sort the rotations, read off the last column. Then watch the transform get inverted back to the original text using nothing but that last column and a table of counts.
Algorithms
Match a pattern right to left, one character at a time, watching the range of matching rows narrow. Then compare the operation count against a naive scan at genome scale.
Algorithms
Sort every suffix of a string and searching becomes binary search, because every occurrence of a pattern lands in one contiguous block. Then see the memory bill that made the Burrows-Wheeler transform necessary.
Algorithms
From every window of consecutive k-mers keep only the smallest. Neighbouring windows usually agree, so the index shrinks fivefold — and two sequences that share a stretch still pick the same ones.
Algorithms
Cut the query into words, look them up, extend only what hits. Then compare the answer with what Smith-Waterman finds on the same pair, and see exactly what the speed cost you.
Algorithms
Drag the read length past the repeat length and watch two assembly methods snap from three contigs to one. Then look at what each costs, and why long reads brought overlap-layout-consensus back from the dead.
Algorithms
Two hidden states, a sequence you can see, and a trellis that recovers the most likely explanation. The same algorithm inside gene finders, HMMER and every basecaller.
Algorithms
Give UPGMA and neighbour-joining the same five distances and watch them build different trees. Then add a fast-evolving lineage and see one of them get the answer wrong without ever saying so.
Algorithms
Every complexity class finishes instantly on a gene. Walk the input up to three billion bases and watch them separate into the ones that finish and the ones that never will.