clarification and accuracy adjustments in refvalidator readme

Author: donaldcampbelljr

bedboss/refgenome_validator/README.md

@@ -1,16 +1,12 @@
 # Reference Genome Predictor and Validator
 
 -----
-### Background
-Original Lab Blog Post: https://lab.databio.org/docs/guide/Reference-genome-predictor-tool/
-
 
 ## Research Project Outline
 
 #### Question
-Original: Can we give a BED file as an input and predict the most likely reference genome assembly associated with the BED file?
 
-Modified: Can we give a BED file as an input and then determine a level of compatibility for different reference genome assemblies?
+Can we give a BED file as an input and then determine a level of compatibility for different reference genome assemblies?
 
 
 #### High Level Execution 
@@ -24,13 +20,11 @@ Modified: Can we give a BED file as an input and then determine a level of compa
    - How many chrom names in BED file that are _not_ in the size file?
 
 3. Compare regions in bed files with excluded ranges/black list regions for each of reference genome.
-
-   - Assumption: during bed file creation, excluded ranges were filtered out. Therefore, if a bed file has more regions in excluded ranges from hg19 vs less from hg38, then the bed file is more likely associated with hg38.
    - Two Tiers: 
-     1. Gaps/Centromeres/Telomeres
-     2. All other Excluded Ranges
+     1. Gaps/Centromeres/Telomeres (can be used to assign tiers for compatibility)
+     2. All other Excluded Ranges (informational only)
 
-4. Once the above are quantified, we can exclude some reference genomes altogether and give probability of compatibility with the remaining.
+4. Once the above are quantified, we can give probability of compatibility with reference genomes.
 
 
 #### Detailed Execution Steps
@@ -39,25 +33,23 @@ Modified: Can we give a BED file as an input and then determine a level of compa
 2. Cache relevant BED files which contain excluded ranges using [BBClient](https://docs.bedbase.org/geniml/tutorials/bbclient/)
 3. Build database for each refgenome assembly excluded ranges, gaps centromeres telomeres using [IGD](https://github.com/databio/gtars/tree/dev_igd) (Use rust implementation if finished, else use C++ implementation).
 4. Query "unknown" BED File against chrom size files and the IGDs (using `igd search`).
-5. Obtain overlap stats for each of the IGDs, rank them based on _least overlaps_.
+5. Obtain overlap stats for each of the IGDs
 6. Run on BED files whose ref genomes are _known_ and calculate accuracy of highest probability compatible ref genome.
 
 
 #### Additional Notes
 - Begin with human bed files and reference genomes for now.
-  - predicting between hg38 and hg19 should be the first (simple) task
-- Prediction happens at a high level for mutually exclusive options,e.g. predicting hg38 vs hg19 as the coordinate systems are different
-  - this mutual exclusivity would also apply to different types of hg38 (UCSC, ensembl, ncbi).
-- Validation occurs at the next level after basic prediction is done.
+- Compatibility assessed via:
   - different/levels of tiers based on cutoffs wrt specificity and sensitivity
   - These tiers are based on a variety of parameters of increasing complexity:
     - name matching of chromosomes
-    - overlaps (chr.sizes, excluded ranges)
+    - size overlaps (chr.sizes)
+    - overlaps with centromeres/telomeres
     - ML (BED embeddings)
     - Bed annotations (text similarity)
 - Future work could involve machine learning for ref genome prediction. However, we will begin with a simple classifier (which may be sufficient).
 - Future work could add annotations/metadata for making the prediction.
 
 #### Software Notes
 
-- Create a validator class that can ingest a BED file as well as a genome_model object
+- Create a validator class that can ingest a BED file as well as a GenomeModel object