diff --git a/data-src/hg38.p2/NCBI/v107/add_gene_biotype_to_gff3.py b/data-src/hg38.p2/NCBI/v107/add_gene_biotype_to_gff3.py
new file mode 100644
index 00000000..0d88cbe8
--- /dev/null
+++ b/data-src/hg38.p2/NCBI/v107/add_gene_biotype_to_gff3.py
@@ -0,0 +1,313 @@
+#!/usr/bin/env python3
+"""
+Add gene_biotype attribute to older NCBI RefSeq GFF3 files that lack it (e.g. v107).
+
+Primary strategy: transfer gene_biotype from a newer RefSeq release using stable
+identifiers (GeneID from Dbxref, then version-stripped accessions), with optional
+coordinate/symbol validation.
+
+Fallback (for unmapped genes): heuristic inference from:
+  - pseudo=true on gene lines -> "pseudogene"
+  - gbkey on transcript lines -> mRNA means "protein_coding",
+    ncRNA uses ncrna_class (lncRNA, miRNA, etc.), precursor_RNA -> "pre_miRNA",
+    misc_RNA -> "misc_RNA"
+  - If a gene has no transcripts, falls back to "misc_RNA"
+
+Usage:
+    python add_gene_biotype_to_gff3.py --reference ref.gff3.gz input.gff3.gz output.gff3.gz
+    python add_gene_biotype_to_gff3.py input.gff3.gz output.gff3.gz  # heuristic only
+"""
+
+import argparse
+import gzip
+import sys
+from collections import defaultdict
+
+
+def open_maybe_gzip(path, mode="rt"):
+    if path.endswith(".gz"):
+        return gzip.open(path, mode)
+    return open(path, mode)
+
+
+def parse_attributes(attr_str):
+    """Parse GFF3 attribute string into a dict."""
+    attrs = {}
+    for item in attr_str.split(";"):
+        item = item.strip()
+        if not item:
+            continue
+        if "=" in item:
+            key, value = item.split("=", 1)
+            attrs[key] = value
+    return attrs
+
+
+def extract_gene_id(dbxref):
+    """Extract GeneID integer from a Dbxref attribute value."""
+    for entry in dbxref.split(","):
+        if entry.startswith("GeneID:"):
+            try:
+                return int(entry[7:])
+            except ValueError:
+                pass
+    return None
+
+
+def strip_version(accession):
+    """Remove version suffix from an accession (e.g. NM_001005484.1 -> NM_001005484)."""
+    if "." in accession:
+        return accession.rsplit(".", 1)[0]
+    return accession
+
+
+def parse_reference_gff3(ref_path):
+    """
+    Parse a reference GFF3 that has gene_biotype on gene lines.
+    Returns:
+        geneid_to_info: {gene_id_int: (biotype, chrom, start, end, symbol)}
+        accession_to_info: {stripped_accession: (biotype, chrom, start, end, symbol)}
+    """
+    geneid_to_info = {}
+    accession_to_info = {}
+
+    print(f"Parsing reference GFF3: {ref_path} ...", file=sys.stderr)
+    count = 0
+    with open_maybe_gzip(ref_path) as f:
+        for line in f:
+            if line.startswith("#"):
+                continue
+            cols = line.rstrip("\n").split("\t")
+            if len(cols) != 9:
+                raise ValueError("Malformed GFF3 line in reference: " + line)
+            # Only care about gene lines
+            if cols[2] not in ("gene", "pseudogene"):
+                continue
+
+            attrs = parse_attributes(cols[8])
+            biotype = attrs.get("gene_biotype")
+            if not biotype:
+                continue
+
+            chrom = cols[0]
+            start = int(cols[3])
+            end = int(cols[4])
+            symbol = attrs.get("gene", attrs.get("Name", ""))
+            info = (biotype, chrom, start, end, symbol)
+
+            # Primary key: GeneID
+            dbxref = attrs.get("Dbxref", "")
+            gene_id = extract_gene_id(dbxref)
+            if gene_id is None:
+                raise ValueError("Reference gene missing GeneID in Dbxref: " + line)
+            geneid_to_info[gene_id] = info
+
+            # Secondary key: version-stripped Name (gene symbol isn't unique, use ID)
+            name = attrs.get("Name", "")
+            if not name:
+                raise ValueError("Reference gene missing Name attribute: " + line)
+            accession_to_info[strip_version(name)] = info
+
+            count += 1
+
+    print(f"  {count} genes parsed from reference ({len(geneid_to_info)} with GeneID).",
+          file=sys.stderr)
+    return geneid_to_info, accession_to_info
+
+
+def lookup_biotype_from_reference(gene_id, name, chrom, start, end, symbol,
+                                   geneid_to_info, accession_to_info,
+                                   max_distance=2_000_000):
+    """
+    Try to find gene_biotype from reference using GeneID (primary) or
+    accession (secondary). Validates chromosome match and coordinate proximity.
+    Returns biotype string or None.
+    """
+    # Try GeneID first
+    if gene_id is not None and gene_id in geneid_to_info:
+        ref_biotype, ref_chrom, ref_start, ref_end, ref_symbol = geneid_to_info[gene_id]
+        # Validate: same chromosome (allow different accession format by comparing last part)
+        if _chrom_compatible(chrom, ref_chrom):
+            return ref_biotype
+        # If chrom doesn't match but GeneID does, still trust it --
+        # could be alt locus mapping differences
+        return ref_biotype
+
+    # Try version-stripped name/accession
+    if name:
+        stripped = strip_version(name)
+        if stripped in accession_to_info:
+            ref_biotype, ref_chrom, ref_start, ref_end, ref_symbol = accession_to_info[stripped]
+            # Validate chromosome and distance
+            if _chrom_compatible(chrom, ref_chrom):
+                distance = min(abs(start - ref_start), abs(end - ref_end))
+                if distance <= max_distance:
+                    return ref_biotype
+            # If only accession matches but coordinates are far, still use it
+            # (genes can move due to assembly patch differences)
+            return ref_biotype
+
+    return None
+
+
+def _chrom_compatible(chrom1, chrom2):
+    """Check if two chromosome identifiers refer to the same sequence."""
+    # Exact match
+    if chrom1 == chrom2:
+        return True
+    # Compare by last component (e.g. NC_000001.11 vs NC_000001.14)
+    base1 = chrom1.rsplit(".", 1)[0] if "." in chrom1 else chrom1
+    base2 = chrom2.rsplit(".", 1)[0] if "." in chrom2 else chrom2
+    return base1 == base2
+
+
+def infer_biotype_from_transcript(gbkey, ncrna_class):
+    """Infer a gene_biotype string from transcript-level attributes."""
+    if gbkey == "mRNA":
+        return "protein_coding"
+    if gbkey == "ncRNA":
+        if ncrna_class:
+            return ncrna_class
+        return "ncRNA"
+    if gbkey == "precursor_RNA":
+        return "pre_miRNA"
+    if gbkey == "rRNA":
+        return "rRNA"
+    if gbkey == "tRNA":
+        return "tRNA"
+    return "misc_RNA"
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Add gene_biotype attribute to NCBI RefSeq GFF3 files that lack it."
+    )
+    parser.add_argument("input", help="Input GFF3 file (optionally gzipped)")
+    parser.add_argument("output", help="Output GFF3 file (gzipped if .gz)")
+    parser.add_argument("--reference", "-r",
+                        help="Reference GFF3 from a newer release (with gene_biotype)", required=True)
+    args = parser.parse_args()
+
+    geneid_to_info, accession_to_info = parse_reference_gff3(args.reference)
+
+    # --- Pass 1: collect transcript info and gene metadata ---
+    tran_info_by_gene = defaultdict(list)
+    gene_is_pseudo = {}
+    gene_has_biotype = {}
+    gene_ids = set()
+    # Store gene metadata for reference lookup
+    gene_meta = {}  # gff_id -> (gene_id_int, name, chrom, start, end, symbol)
+
+    print(f"Pass 1: scanning {args.input} ...", file=sys.stderr)
+    with open_maybe_gzip(args.input) as f:
+        for line in f:
+            line = line.rstrip("\n")
+            if not line or line.startswith("#"):
+                continue
+
+            cols = line.split("\t")
+            if len(cols) != 9:
+                continue
+
+            attrs = parse_attributes(cols[8])
+            gff_id = attrs.get("ID", "")
+
+            if cols[2] == "gene" or cols[2] == "pseudogene" or gff_id.startswith("gene"):
+                gene_ids.add(gff_id)
+                gene_is_pseudo[gff_id] = (
+                    attrs.get("pseudo", "") == "true" or cols[2] == "pseudogene"
+                )
+                gene_has_biotype[gff_id] = "gene_biotype" in attrs
+
+                # Extract metadata for reference lookup
+                dbxref = attrs.get("Dbxref", "")
+                gene_id_int = extract_gene_id(dbxref)
+                name = attrs.get("Name", "")
+                symbol = attrs.get("gene", name)
+                chrom = cols[0]
+                start = int(cols[3])
+                end = int(cols[4])
+                gene_meta[gff_id] = (gene_id_int, name, chrom, start, end, symbol)
+
+            elif gff_id.startswith("rna"):
+                parent = attrs.get("Parent", "")
+                if parent in gene_ids or parent.startswith("gene"):
+                    gbkey = attrs.get("gbkey", "")
+                    ncrna_class = attrs.get("ncrna_class", "")
+                    tran_info_by_gene[parent].append((gbkey, ncrna_class))
+
+    # --- Compute biotypes ---
+    gene_biotype = {}
+    stats = {"reference": 0, "heuristic": 0, "already": 0, "id_not_found": 0, "name_not_found": 0}
+
+    for gid in gene_ids:
+        if gene_has_biotype[gid]:
+            stats["already"] += 1
+            continue
+
+        # Try reference lookup
+        gene_id_int, name, chrom, start, end, symbol = gene_meta[gid]
+
+        if gene_id_int not in geneid_to_info:
+            stats["id_not_found"] += 1
+        elif strip_version(name) not in accession_to_info:
+            stats["name_not_found"] += 1
+
+
+        biotype = lookup_biotype_from_reference(
+            gene_id_int, name, chrom, start, end, symbol,
+            geneid_to_info, accession_to_info
+        )
+        if biotype:
+            stats["reference"] += 1
+
+        # Fallback to heuristic
+        if biotype is None:
+            if gene_is_pseudo[gid]:
+                biotype = "pseudogene"
+            elif tran_info_by_gene[gid]:
+                gbkey, ncrna_class = tran_info_by_gene[gid][0]
+                biotype = infer_biotype_from_transcript(gbkey, ncrna_class)
+            else:
+                biotype = "misc_RNA"
+            stats["heuristic"] += 1
+
+        gene_biotype[gid] = biotype
+
+    print(
+        f"  {len(gene_ids)} genes total, {stats['already']} already have gene_biotype, "
+        f"{stats['reference']} mapped from reference, {stats['heuristic']} inferred by heuristic.\n"
+        f"  Reference mapping failures: {stats['id_not_found']} with GeneID not found, "
+        f"{stats['name_not_found']} with Name not found.",
+        file=sys.stderr,
+    )
+
+    # --- Pass 2: write output with gene_biotype injected ---
+    print(f"Pass 2: writing {args.output} ...", file=sys.stderr)
+    with open_maybe_gzip(args.input) as fin, open_maybe_gzip(args.output, "wt") as fout:
+        for line in fin:
+            line = line.rstrip("\n")
+            if not line or line.startswith("#"):
+                fout.write(line + "\n")
+                continue
+
+            cols = line.split("\t")
+            if len(cols) != 9:
+                fout.write(line + "\n")
+                continue
+
+            attrs = parse_attributes(cols[8])
+            gff_id = attrs.get("ID", "")
+
+            if gff_id in gene_biotype:
+                cols[8] = cols[8].rstrip(";") + f";gene_biotype={gene_biotype[gff_id]}"
+                fout.write("\t".join(cols) + "\n")
+            else:
+                fout.write(line + "\n")
+
+    print("Done.", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    main()
+
diff --git a/data-src/hg38.p2/NCBI/v107/fetch.sh b/data-src/hg38.p2/NCBI/v107/fetch.sh
new file mode 100644
index 00000000..360f59ba
--- /dev/null
+++ b/data-src/hg38.p2/NCBI/v107/fetch.sh
@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+wget -O v107.src.gff3.gz https://ftp.ncbi.nlm.nih.gov/genomes/archive/old_refseq/Homo_sapiens/ARCHIVE/ANNOTATION_RELEASE.107/GFF/ref_GRCh38.p2_top_level.gff3.gz
+
+# Download a recent RefSeq release as reference for gene_biotype transfer
+wget -O ref_latest.gff3.gz https://ftp.ncbi.nlm.nih.gov/genomes/all/annotation_releases/9606/GCF_000001405.40-RS_2024_08/GCF_000001405.40_GRCh38.p14_genomic.gff.gz
+
+# The v107 gff is missing gene_biotype; transfer from the newer release, with heuristic fallback
+python add_gene_biotype_to_gff3.py --reference ref_latest.gff3.gz v107.src.gff3.gz v107.tmp.gff3.gz
+
+# NT_187507.1 isn't in hg38.p12.chromAlias.txt, which we are using for hg38.p2.
+# ok to remove it because NCBI says:
+# > Record suppressed. This RefSeq record was removed because the sequence was determined to have originated from Chinese hamster.
+# https://www.ncbi.nlm.nih.gov/nuccore/NT_187507.1?report=genbank
+
+# doing extra work of unzipping and gzipping again but saves code duplication and only takes a few seconds longer
+zgrep -v NT_187507.1 v107.tmp.gff3.gz | gzip > v107.gff3.gz
+rm v107.tmp.gff3.gz
diff --git a/data-src/hg38.p2/assembly/fetch.sh b/data-src/hg38.p2/assembly/fetch.sh
new file mode 100644
index 00000000..53a7c170
--- /dev/null
+++ b/data-src/hg38.p2/assembly/fetch.sh
@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+
+wget https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_22/GRCh38.p2.genome.fa.gz
+faToTwoBit GRCh38.p2.genome.fa.gz hg38.p2.2bit
+twoBitInfo hg38.p2.2bit stdout | sort -k2rn > hg38.p2.chrom.sizes
+
+# need chromAlias for ncbi, using hg38.p12.chromAlias.txt
+wget -O hg38.p2.chromAlias.txt https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/p12/hg38.p12.chromAlias.txt