Difference between revisions of "DOCK GA Development Goals"
From Rizzo_Lab
(→To Do List) |
BrockBoysan (talk | contribs) |
||
Line 37: | Line 37: | ||
|- | |- | ||
| metropolis selection for tournament/roulette || conf_gen_ga.cpp || BTB || no | | metropolis selection for tournament/roulette || conf_gen_ga.cpp || BTB || no | ||
+ | |- | ||
+ | | Selection of fragments for mutation based on Fragment Score || conf_gen_ga.cpp || BTB || no | ||
|- | |- | ||
|} | |} |
Revision as of 12:38, 8 January 2024
Tasks | src | Owner | Complete? |
---|---|---|---|
Horizontal pruning issue not catching chemically identical molecules | LEP | no | |
Fix bug that collapsed atom coordinates | everywhere? nowhere? somewhere. | LEP | YES MUAHAHAHAHA! |
Added Delimeter header | conf_gen_ga | LEP | Yes |
Fix xover only feature | conf_gen_ga | LEP | Yes |
Put in error messages for mut_rate > 1 | conf_gen_ga | LEP | Yes |
Manual user-defined mutation type | conf_gen_ga | LEP | Yes |
Remove check only option | conf_gen_ga | LEP | Yes |
Add single molecule evolution in testcase in install dir. | install/test/genetic | LEP | yes |
Add leading 0's to xover output filenames | conf_gen_ga.cpp | JDB | no |
DNM replacement unable to build list? | conf_gen_ga.cpp | JDB | no |
Multi-layer replacement for Amides | conf_gen_ga.cpp | JDB | no |
Compute delta slope of fitness score | congen_ga.cpp | BTB | no |
slow down molecular evolution so there are less drastic canges between each successive generation | congen_ga.cpp | BTB | no |
bring in new parents (e.g. from a pool of molecules) based on convergence | confgen_ga.cpp | BTB | no |
user defined point vs on-the-fly convergence | confgen_ga.cpp | BTB | no |
metropolis selection for tournament/roulette | conf_gen_ga.cpp | BTB | no |
Selection of fragments for mutation based on Fragment Score | conf_gen_ga.cpp | BTB | no |
To Do List
- tanimoto coefficient percent change - might be inaccurate due to tan coef behavior
- Rotatable bond changes (???)
- Limit number of aromatic rings.
- -xover (guided based on score) - Good v Good ; Bad v Good ; Bad v Bad THIS
- Nonexhaustive xover (pick subset of xover based on probability)
- Nonexhaustive xover for each pair of parents - have a set number of bonds for xover rather than trying all of them.
- Crossover on multiple points simultaneously (2-3 crossovers on a given pair to make 3+ children rather than 2+).
- Adaptive maintenance of ensemble based on convergence (extinction, delta max, etc.).
- Stop convergence.
- Mutations-
- Adaptive mutation rate - change rate of mutation based on some internal criteria.
- Pick location of mutation based on some internal criteria.
- Pick mutation type based on ensemble behavior.
- If molecules are too large, boost deletion (useful for elitism).
- If molecules are too small, boost additive mutations.
- If molecules are too similar, boost replacements and substitutions.
- mutation type selection based on probability vs ensemble
- Complete x # y mutation so far so less prevalent etc
- Note: 3 layer substitutions probably aren't going to work.
- 2-layer replacements.
- fitness-
- turn on and off niching adaptive/extinction
- reduce boost of fragments and all poor mols with niching
- pareto/mulitobjective ga
- selection-
- Adaptively keep differing #'s of parents and children based on some internal criteria (similarity?)
- extinction-
- which molecules are best-
- best first pruning - now uses descriptor score even if niching ned to delta to fitness/niching when used
- geometric diversity using Hingarian and Tan pruning
- Determine whether the dummy vdw file is necessary for all DN (including GA) or just DN.
- Choose crossover pairs based on measure of similarity (boost crossovers between dissimilar molecules).
Known Bugs
- Molecules Processed bug (dock.cpp)
- verbose mol stats (amber typer)
- molecule being renamed when going into repl even tho it's the same molecule