martinjvickers / alfsc_old Goto Github PK

A bad read is one where there are Zero K-Mer's. This can occur for a couple of reasons but the symptom is that any Iupac characters other than an AGCT gets converted to an N. If the whole input contig does not contain AGCTs, then everything gets converted to an N and the you have various divide by zero errors. In order to deal with this I need to;

• Check the number of returned counts.
• I need to start storing the contig name along with the counts so that I no longer need to rely on the order of the users input files. This is especially important when skipping reads.
• I need to tell the user that this is happening.

Need to ensure mutex's are in place too.

I want this to play nice with memory, e.g. either a --mem flag, or maybe it loads things into memory and if it is too much for the machine, reverts to a low memory state?

This is very important for very large kmer sizes e.g. 4^9 isn't that unreasonable but is massive.

• In order to do this, I need to reimplement all the calculations and distances to be okay with missing data, e.g. reverting to zero if there is no count from say a hashtable
• The order also needs to be preserved for markov doesn't it?
• If I can manage this, it will reduce the amount of memory used.

At the moment, if you want to use the RAM option for the precompute of the reference, it runs sequentially. If this were to be multithreaded, it would be very useful.