Merge pull request #29 from TravisWheelerLab/dev

unbump versions

Author: jackroddy

.github/workflows/release.yml

@@ -107,7 +107,7 @@ jobs:
     needs: tag-bin
     strategy:
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-13]
+        os: [ubuntu-latest, windows-latest, macos-latest]
         arch: [x86_64, aarch64]
         exclude:
           - os: windows-latest
@@ -120,7 +120,7 @@ jobs:
                   matrix.arch == 'x86_64' && 'x86_64-unknown-linux-gnu' ||
                   matrix.arch == 'aarch64' && 'aarch64-unknown-linux-gnu'
                 ) || matrix.os == 'windows-latest' && 'x86_64-pc-windows-msvc' ||
-                matrix.os == 'macos-13' && (
+                matrix.os == 'macos-latest' && (
                   matrix.arch == 'x86_64' && 'x86_64-apple-darwin' ||
                   matrix.arch == 'aarch64' && 'aarch64-apple-darwin'
                 ) }}

libnail/CHANGELOG.md

@@ -27,14 +27,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - added function `Profile::from_blosum62_and_seq()`
 
 ### Changed
-- removed transition index constants under `Profile` namespace
+- Profile::from_blosum_62_and_seq() now estimates tau with a closed-form solution
 - renamed `Profile.forward_tau` and `Profile.forward_lambda` to `fwd_tau` and `fwd_lambda`
 - renamed `Profile.consensus_sequence_bytes_utf8` to `consensus_seq_bytes_utf8`
 - moved `Alphabet` and`AminoAcid` to mod `alphabet`
 - split self mutating functions from trait `VecMath` into new trait `VecMathMut`
 - struct `Seed` now has `prf`, `seq`, `e_value` fields (probably temporary)
 
 ### Removed
+- removed transition index constants under `Profile` namespace
 - removed `Hmm` struct
 - removed function `Profile::new()`
 

libnail/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "libnail"
-version = "0.5.0"
+version = "0.4.0"
 authors = ["Jack Roddy <jack.w.roddy@gmail.com>"]
 edition = "2021"
 license = "BSD-3-Clause"

libnail/src/structs/profile.rs

@@ -788,12 +788,24 @@ impl Profile {
             })
             .sum::<f32>();
 
-        prf.fwd_tau(0.0);
+        // note: we could get also get a pretty good approximation for
+        //       lambda by simply using the product of the length of the
+        //       sequence and the mean relative entropy of blosum62
         prf.fwd_lambda(std::f32::consts::LN_2 + 1.44 / total_entropy);
 
-        let mut prf = prf.build()?;
-        prf.calibrate_tau(200, 100, 0.04);
-        Ok(prf)
+        // tau
+        // note: this all comes from some experimentation fitting ~200k
+        //       sequences to learn a closed form solution for tau
+        const KL_BLOSUM62: f32 = 0.589;
+        const MAGIC_TAU_1: f32 = 0.662;
+        const MAGIC_TAU_2: f32 = 1.317;
+        const MAGIC_TAU_3: f32 = 2.147;
+        let tau = -((-MAGIC_TAU_1 * KL_BLOSUM62) + (MAGIC_TAU_2 * (seq.length as f32).ln())
+            - MAGIC_TAU_3);
+
+        prf.fwd_tau(tau);
+
+        prf.build()
     }
 
     pub fn relative_entropy(&self) -> f32 {

nail/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "nail"
-version = "0.5.0"
+version = "0.4.0"
 authors = ["Jack Roddy <jack.w.roddy@gmail.com>"]
 edition = "2021"
 license = "BSD-3-Clause"
@@ -20,7 +20,7 @@ jemalloc = ["jemallocator"]
 [dependencies]
 jemallocator = { version = "0.3", optional = true }
 clap = { version = "4.5.23", features = ["derive", "wrap_help"] }
-libnail = { path = "../libnail", version = "0.5.0" }
+libnail = { path = "../libnail", version = "0.4.0" }
 anyhow = "1.0.66"
 thiserror = "1.0.37"
 rayon = "1.7.0"