Fix/off by one in search chains

[niyarin]

This PR fixes an issue where the search range of in-block? was off by one. Similar to search-overlap-block, it needs to search one position further back.

[alumi]

Thank you for the PR!

I looked into this, and it seems the root cause is that we’re mixing half-open and closed intervals.
The intervals/find-overlap-intervals API expects 1-based closed intervals, but the current implementation of cumsum-chain does not fully convert the chain file format (0-based half-open) into 1-based closed intervals.
Strangely, changing (when-let [m (first (intervals/find-overlap-intervals data nil (inc pos) (+ pos 2)))] to (when-let [m (first (intervals/find-overlap-intervals data nil (inc pos) (inc pos)))] still passes the tests.

Since most part of cljam uses closed intervals, I think it's more straightforward to update cumsum-chain so that it produces closed intervals.
Does this approach sound reasonable to you?

diff --git src/varity/chain.clj src/varity/chain.clj
index d46d26e..ce563c3 100644
--- src/varity/chain.clj
+++ src/varity/chain.clj
@@ -80,10 +80,10 @@
       (persistent! results)
       (recur (->> (assoc d :t-start curr-t-start
                          :q-start curr-q-start
-                         :t-end (+ curr-t-start (:size d))
-                         :q-end (+ curr-q-start (:size d))
+                         :t-end (dec (+ curr-t-start (:size d)))
+                         :q-end (dec (+ curr-q-start (:size d)))
                          :start curr-t-start
-                         :end (+ curr-t-start (:size d)))
+                         :end (dec (+ curr-t-start (:size d))))
                   (conj! results))
              (first r)
              (next r)
@@ -114,9 +114,8 @@
 (defn- in-block?
   [pos {:keys [data header] :as chain}]
   (when (<= (inc (:t-start header)) pos (:t-end header))
-    (when-let [m (first (intervals/find-overlap-intervals data nil (inc pos) (+ pos 2)))]
-      (when (<= (:t-start m) pos (dec (+ (:t-start m) (:size m))))
-        (assoc chain :in-block m)))))
+    (when-let [m (first (intervals/find-overlap-intervals data nil pos pos))]
+      (assoc chain :in-block m))))

 (def ^:private normalize-chr (memoize normalize-chromosome-key))

@@ -138,4 +137,4 @@
 (defn search-overlap-blocks
   "Calculates a list of blocks that overlap the given interval."
   [start end blocks-idx]
-  (intervals/find-overlap-intervals blocks-idx nil (inc start) end))
+  (intervals/find-overlap-intervals blocks-idx nil start end))
diff --git test/varity/chain_test.clj test/varity/chain_test.clj
index cb08756..5e577be 100644
--- test/varity/chain_test.clj
+++ test/varity/chain_test.clj
@@ -68,7 +68,7 @@
     15 17 [{:t-start 14, :t-end 16}]
     6 7 [{:t-start 6, :t-end 10}]
     4 4 [{:t-start 2, :t-end 5}]
-    5 5 []
+    5 5 [{:t-start 2, :t-end 5}]
     6 6 [{:t-start 6, :t-end 10}]
     1 1 []
     30 31 []))

[niyarin]

Thank you. Your point seems to be correct.
The test you proposed also appears to be correct; this is consistent with the behavior before introducing nc-list.
I will change the way interval endpoints are handled to closed intervals.

[alumi]

@niyarin Thank you for updating the PR! Looks good👍
Would you squash the last commit into the second one?