introgression_plan.rtf

{\rtf1\ansi\ansicpg1252\cocoartf1404\cocoasubrtf470
{\fonttbl\f0\fswiss\fcharset0 Helvetica;}
{\colortbl;\red255\green255\blue255;}
\margl1440\margr1440\vieww11160\viewh14280\viewkind0
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0

\f0\b\fs24 \cf0 Simulations
\b0 \
For the coalescent simulations, I need to choose reasonable model parameters. Some of these I\'92ve gotten from estimates in the literature, some of them from comparing the properties of the simulated sequences to real sequences, and for some I just need to try a range of values.\
    \'95 from literature: mutation rate, recombination rate\
    \'95 from comparing to observed sequence divergence within and between cerevisiae/paradoxus: divergence time\
    \'95 vary: effective population size, outcrossing rate, migration rate\
\
From simulations without migration (or just using math), I can evaluate how likely ILS is to occur. It\'92s not likely, but I should maybe also demonstrate how different the parameters (specifically effective population size and mutation rate) would have to be for ILS to become an issue.\
\
Then, adding in migration to the simulations, I can evaluate my method\'92s performance and compare it to the phylo-hmm method. Things to compare:\
    \'95 runtime\
    \'95 power\
    \'95 false positive rate\
    \'95 overlap in predictions\
\

\b 100-genomes data\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0

\b0 \cf0 Run HMM using S288c and CBS432 as references.\
Maybe also run with a different cerevisiae and/or paradoxus reference.\
Summarize properties of introgressed regions:\
    \'95 how many?\
    \'95 how large?\
    \'95 how many genes?\
    \'95 correlation with polymorphism rate across the genome\
    \'95 correlation with recombination rate\
\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0

\b \cf0 Figures
\b0 \
1. theory/ILS/comparison of methods on simulated data\
2. overall summary of introgressed regions (chromosome plots)\
3. something about region sizes/breakpoints\
4. specific interesting introgressed genes/regions
\b \
\
\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0

\b0 \cf0 figure 1\
- simulate data with variety of migration rates (more recent and over all time)\
- 50000bp, 100 sims \
- vary initial parameters, see how results (tpr/fpr, final hmm param vals) change\
- train on more sequences? then predict without training?\
- something about how close it gets to breakpoints\
\
for fixed everything else (mig = 1e-10):\
	expected tract lengths: \
	expected num tracts: \
\
100 x 1, 100 x 10, 1000 x 20\
1000 x 1, 1000 x 5, 1000 x 10\
10000 x 1, 10000 x 2, 10000 x 3\
\
for fixed expected tract length (1000) and num tracts (2):\
	migration rate: 0,  2.5e-11, 5e-11, 1e-10, 2.5e-10, 5e-10, 1e-9, 2.5e-9\
\
is it okay to have bayanus with no expected introgressed? or does that break the model?\
\
ideally would run predictions varying initial hmm parameters on same simulated sequences\
\'97> should be set up in a way where you specifiy simulation parameters and give that a tag, then you specify prediction parameters in another file with associated simulated sequence tag\
\
\
\
simulation plots\
- tpr, fpr as function of amount of introgression/migration\
- robustness to choice of starting parameters\
	> vary number/length of expected introgressions\
	> look at how much initial and final parameters change\
	> look at tpr, fpr\
\
\
\
\
qtls that influence variance in gene expression\
10x\
parental strains sequencing \
\
\
roc curve for hmms?\
scatterplot for my method and phylonet hmm\
\
vary time of migration, do as single point\
show that priors actually matter by setting them to something stupid\
\
run again on empirical data\
amounts of introgression in paralogs vs single copy genes\
fsa on specific regions that look interesting\
\
bed file for each strain of introgression calls\
\
distribution of number of genes/total sequence lengths/median tract length per strain\
\
comparing calls to the 100 genomes paper\
\
anything we can infer\
\
how much does using a different reference change results? unique vs shared introgressions\
\
how much do strains differ in divergence to paradoxus locally\
\
genome research, plos genetics, mbe, pnas (?)\
\
number of nonsynonymous changes\
\
high frequency introgressions\
\
distribution of introgression length and frequency\
\
gene ontology \
\
coding vs noncoding regions\
\
phylogeny based on pairwise sharing of sequences compared to genome-wide phylogeny\
\
\
\
outline: using simulations to evaluate my method\
- simulating useful sequences\
- evaluating tpr/fpr etc\
- evaluating robustness\
- comparison to phylonet hmm\
- other things to look at\
\
\
simulating useful sequences\
- two species\
- range of migration rates, hoping to get intermediate level\
- migration through all time -> vast majority of sequences not detectable\
- migration more recently only (average over a range of recent times)\
- calculated results very dependent on what kinds of sequences you\'92re actually including\
- summary info about sequences i\'92m actually using\
	- varying migration rate zero to high\
	- varying initial parameters by varying expected number of tracts/length of tracts (robustness)\
\
performance of my method\
- tpr\
- fpr\
- hmm params\
- some things about tract lengths/boundaries\
\
comparison to phylonet hmm\
- description of method\
- sequences used for comparison\
- tpr/fpr\
- examples of predictions\
\
\uc0\u8232 \
other things\
- lengths/boundaries\
- simulate points of migration\
- look at robustness at different levels of introgression\
\
\
\
TODO:\
add slide explain initial hmm parameter values\
look at robustness for different migration rate (ie somewhere where transition probabilites aren\'92t always going in the same direction(?))\
maybe take out simulating sequences intro except actual values->do all background on board\
\
\
why doesn\'92t initial par->cer prob depend on number of tracts, but cer->par does\
sim_predict line 366 fix this\
\
\
\
power/tpr as function of fdr\
table of comparing methods, fpr for all different parameter sets\
\
fdr = tp/pp}