Merge pull request #32 from EvanKomp/dev

Dev

Author: EvanKomp

docs/_build/latex/aide.pdf

@@ -7,6 +7,8 @@ See below for the cost it takes to run each tool. This test was run with a Dual
 
 The test system was a GFP (238) amino acids, MSA depth (when applicable) was 201. Times measure the total time to fit the model (when applicable) and run prediction on 50 variants. Missing values are either because the core model does not support that type of prediction or because AIDE's wrapper does not support it.
 
+NOTE: The cost of some of these (*) is significantly impacted by hyperparameters.
+
 ## Zero shot predictors
 
 | Model Name | Marginal Method | GPU Total Time (s) | CPU Total Time (s) |
@@ -23,7 +25,7 @@ The test system was a GFP (238) amino acids, MSA depth (when applicable) was 201
 | SaProtLikelihoodWrapper | mutant_marginal | 7.326 | 220.906 |
 | SaProtLikelihoodWrapper | masked_marginal | 14.814 | 429.626 |
 | VESPAWrapper | - | 244.852 | - |
-| EVEWrapper | - | 925.930 | - |
+| EVEWrapper * | - | 925.930 | - |
 | SSEmbWrapper | - | 192.999 | - |
 
 ## Embedders

docs/user_guide/resource_test.md

@@ -2,7 +2,7 @@
 title: Roadmap
 ---
 
-# Roadmaps
+# Roadmap
 
 ## Additional predictors from maintainers
 __Quarter 2, 2026__
@@ -15,6 +15,7 @@ We have the following predictors planned to be wrapped:
   - Type 3 dependencies (sub environment)
 2. NOMELT (as a zero shot predictor)
   - Temperatute specific scores
+  - Type 3 dependencies
 3. ESM3 (Zero shot prediction and embedder)
   - Structure aware if available
   - For now, the annotation data mode will not be considered. This will require an additional attribute of the ProteinSequence class (in addition to sequence, structure, msa data types already supported).
@@ -31,6 +32,4 @@ The component specification and software engineering exercise conducted in AIDE
     - ProteinMPNN
     - NOMELT
     - Tranception
-    - 
-- Score optimizers, eg. BADASS (already included in the package) which use a scoring function (AIDE predictor) to bias generation.
-    - This type may even be able to be wrapped with Conditional Generators for tandem generation and filtering.
+- Score optimizers (black box or maybe gradient aware), eg. BADASS (already included in the package) which use a scoring function (AIDE predictor) to bias generation.