• Memory efficient, uses compressed data structures
• Very modular, each component can be run independently
• Emphasis on accuracy
• Most of the CPU time is in steps that can be parallelized and merged
• Tries to excel at substring coverage

• Current pipeline is different than the original paper, documentation is unclear or scattered
• Minimum of 20x coverage, recommended 40x coverage.
• Reads must be 100bp or greater
• Key parameters: overlap size, k-mer size for error correction

• Relies on BWT (Burrows Wheeler Transform), which allows for reversible compression and rearranges the string into runs of similar characters
• FM-Index (Full-text Index in Minute space) - scales with the size of the input alphabet
• Only some fraction of the last column of the BWT is stored
• First column does not store nucleotides, merely pointers to where each nucleotide would start

1. Construct FM Index
2. Merge paths to reduce graph size
3. Re-index
4. Build string graph

The assembly algorithm queries the FM Index and follows (unambiguous) paths where one k-mer maps to a given end nucleotide and condenses these paths to a single read.

• Remove duplicate reads, index with l-mers and then check reads with shared l-mers for longer overlaps
• Builds edges and labels them with the non-matching sequence from two overlapping k-mers and removes transitive edges

• For a pair of nodes with multiple walks, choose one to remain
• Compare other walks and if they are similar enough (95%) to the chosen walk, remove them

• Create a potential order of contigs, removing uncertain or repetitive ones
• Standalone module, uses k-mer search on gaps to find paths

• Successfully run sga preprocess, sga index, sga correct, sga index (after correction)

• Adjust error correction and assembly
• Finish running full pipeline on one library
• Re-run completed pipeline using all libraries