1 GREYLAG TODO LIST -*-outline-*-
4 ==============================================================================
8 1. Replace SEQUEST at SIMR with something at least as good.
9 2. Do better than SEQUEST for things that SIMR cares about.
10 3. Showcase Python w/C-ish inner loop code implementation strategy.
11 4. Try to take the best ideas from other similar programs.
12 5. Greylag as a pedagogical artifact and foundation for further
15 ==============================================================================
20 * Good first impression
22 * Handles at least LCQ input
23 * Nonspecific cleavage only?
24 * Generates SQT output, usable in our pipeline, at least for non-N15 runs
25 * Decent performance/efficiency on our clusters
29 * Basic public website/source release/git archive
30 * Documentation (asciidoc/man pages)
36 * MINI-GOAL: Get something working that can be tested against MM/SEQUEST
37 * MINI-GOAL: basic greylag-process usable on our cluster (no mods)
40 * Demo runs! (vs SEQUEST)
41 ** observe memory use!
44 ** ASFP (multiple mods)
45 ** vs myrimatch, if run completes
48 ** What it does and doesn't yet do
51 * Change the way mod limit works? ({1,4} feature?)
53 * Why aren't scores a little more stable? (fast-math?)
55 * add A1S2D1C3A1S2D1F4 marking form?
59 * Evaluate performance differences vs SEQUEST/MM/Xtandem?
61 * Design and implement greylag master process (work manifests?)
63 * Compare greylag/SEQUEST/MM on test-myrimatch example (non-specific)
64 ** look at MM (whole file)
65 ** Note: SEQUEST parent tolerance differs
68 ** look at callgrind output
70 * Further test case updates/adds
74 * Verify that memory usage no longer a problem
75 ** maybe avoid spectrum name copies
76 ** maybe avoid locus name copies
77 ** instead of copying db sequences, use Python's?
80 = M1 =========================================================================
84 * do a cg-admin-rewritehist before publishing git archive?
89 = M2 =========================================================================
91 * Implement MM smart +3 model?
94 * Try to generate a valid MyriMatch (bombs on boost random assertion)
96 * Try the MM precursor mass adjustment--much improvement? even a good idea?
98 * Test tolerance monotonicity
102 ** Design similar statistical evaluation
103 ** Look at what we do here (paper)
106 * Add isotope jitter feature, for Orbitrap.
108 xtandem considers one C13 if MH>1000, and one/two C13 if MH>1500. Should we
109 try to predict this based on the peptide sequence? MH probably close
110 enough. What does MyriMatch do?
113 * Implement MyriMatch charge-calling algorithm?
115 * Implement MyriMatch deisotoping?
118 * Look more closely at specific MM vs GL match diffs.
121 * Figure out how to handle multiple residue mods (delta, isotope, etc)
123 * Clean up [,] (N/C-terminal) mass regime calculations
125 * Pass through the C++ code looking for counts that could conceivably overflow
126 ** Fix or add assertions
128 * Add duplicate peptide masking optimization
129 ** This will obviate the need to detect identical best matches at search time?
130 ** Fix redundant peptide reporting
133 * Time a real mod search vs SEQUEST (and xtandem?), is time reasonable?
134 MAKE SURE PARAMS ARE COMPARABLE! Ballpark correctness?
136 * Create direct DTASelect.txt output?
137 (This seems to be sufficient to support most or all DTASelect output.)
139 * Make a tool to compare greylag vs SEQUEST results by spectrum. Want gross
140 statistics--how many id's are the same, different, missing, etc. For each
141 spectrum, want to see what each program did, and how many times the assigned
142 locus was otherwise id'ed.
144 * Investigate identification differences between greylag and SEQUEST.
146 * Careful timing and correctness check for
147 /n/proteomics/mkc/HsProA-Control_S100_Ti_1_H_2006-03-03_wSHUFFLED-greylag
149 * Design and implement tracing of mass regime/PCA/fixed and non-fixed
150 deltas/etc into output file. Try to stay compatible with xtandem.
152 * PPM error tolerances (MyriMatch doesn't implement this?)
154 * Make --estimate work correctly over cluster. (Currently takes 6 hours to
155 estimate 60--is this worthwhile? Could we simply estimate one bag and
156 multiply by the number of bags??)
158 * Better shuffling than current model.
160 * Useful to scale fragment tolerance by charge, too?
162 * Have --estimate generate a spectrum work count file (*.est?) that can be
163 used by --part-split to generate evenly sized parts. (Check that file is
164 newer than params file and ms2 file arguments, and that all ms2 file
165 arguments were estimated.)
167 * Maybe --part-split should generate a downramp of sizes? It definitely
168 should take into account spectra filtered out (== no work), but this
169 requires reading all spectra before splitting (which takes more time).
171 * Fix "cleavage C-terminal mass change" issue. Should this be interpreted as
172 MONO, ! (first fragment regime), or by regime. Look for similar problems
175 * Make static '[' mod exclude PCA mods.
177 * Code cleanup, especially in new Python code. Maybe put some stuff in
178 classes. Could split into multiple source files.
180 * Mine OMSSA and myrimatch for ideas. Look again at X!Tandem and SEQUEST
183 * Need tool to compare two runs, for regression testing purposes.
185 * Add refinement. (like xtandem?)
187 * Advanced refinement ideas. For example, only search a locus for a hit with N
188 mods if we got a hit for it with 0..N-1 mods (or maybe 0..N-2?). Or, only
189 search a locus non-tryptically (or semi-tryptically) if we got a tryptic hit
192 ** Investigate current SEQUEST search results to see if this looks feasible.
194 * Think about ways to get more id's per hour of processing time.
196 * Try to adapt to instrument accuracy. Maybe start with a narrow parent mass
197 range and adaptively widen it.
199 * Profiling to find slow spots, and for correctness?
201 * Heavy optimization on inner loop. Try running from both ends simultaneously.
203 * Should we try to guarantee that searching is equivalent for all mass regimes,
204 to make comparisons more valid?
206 * Eval speedup: make intensity information integer, or otherwise store it in
207 log form so we can add instead of multiplying?
209 * Try to switch FP code to use integers instead?
211 * Rigorously check all values coming in from Python (at least by assert).
213 * We can now pre-build a peptide index if we want to. The main utility of this
214 is that it would allow us to avoid searching a spectrum against the same
215 peptide multiple times (saving perhaps 30% runtime for one real database).
216 Alternatively, maybe we could just generate a description of peptides to be
219 * Look at moving C++ code to C+ctypes, or maybe pyrex?
221 * Incrementalize the whole program. Want to be able to take an existing run
222 and spend more time on it to get more results, possibly concentrating on a
223 particular kind of modification.
225 * Try to figure out whether SEQUEST is really searching everything, or whether
226 it gives up in certain cases like X!Tandem does.
228 * Isotope S34 and C13 are common (4%, 1%). Is there a good way to look for
229 them? We could look for singleton occurrences pretty cheaply using a delta
232 * Splitting idea: Rather than having all parts be equal, maybe its better for
233 the parts processed first to be bigger, with smallest parts processed last,
234 so that they can fill in the final gaps (leading all processors to finish at
235 about the same time). What should the split curve look like? Linear, but
236 what slope? No split should be smaller than one spectrum (after
239 * Possible generation optimization: Figure out the maximum number of mods,
240 which would be the number that could be added to the smallest, lightest
241 peptide without exceeding the mass of the largest spectrum parent mass.
242 Probably not worth doing? Similarly, if we know all bags of size N are too
243 large, and all deltas are positive, we can skip larger bags.
245 * Can we make the parts restartable? If so, maybe this could be used to load
246 balance, recover from crashes, etc.
248 * Is NOTHROW faster or slower?
250 * If we see a good hit for a spectrum, we could try to see if there's an
251 identifiable tag. If so, could restrict further searching to peptides with
257 * What is this C+57 mod called? Carboxyamidomethyl? +C2H3ON!
259 * Is there anything we can do with neutral losses?
261 * test spectrum synthesis
263 * test semi-tryptic cleavage
265 * double-check handling of FP arithmetic using epsilons (no ==, no strict <)