Exception in thread main java.lang.OutOfMemoryError" GATK HaplotypeCaller fails: how to fix

• The traceback says OutOfMemoryError and Java heap space during a GATK HaplotypeCaller run. - The same BAM called variants fine in a smaller test region or on a machine with a larger heap. - The GATK log shows a default or small heap even though the machine has free memory.

Confirm the heap GATK actually started with. Check the GATK log header for the Java command line and the max heap value. Expected: the log shows a small default heap, often far below the memory free on the machine.

Rerun with an explicit larger heap. Add the java options setting to the HaplotypeCaller command, for example a heap of 8 gigabytes on a machine that has that memory free, keeping the rest of the command unchanged. Success check: the log header now shows the larger max heap you set.

Test on one small interval first. Run the same command limited to a single short interval or a small chromosome region. Expected: the test run finishes and writes a valid VCF instead of ending with OutOfMemoryError.

Split large jobs by interval. If whole genome runs still exhaust a reasonable heap, run HaplotypeCaller per interval in parallel and merge the resulting VCF files afterwards. Success check: each interval job completes on its own heap and the merged VCF contains records from every interval.

Check for an oversized input problem. Look for an unindexed BAM, a BAM with an unusually deep region, or a missing interval list that makes the tool load more than intended. Expected: the input BAM has an index file beside it and the run is scoped to the intended genome.

Resolved from the public problem: https://vectle.com/posts/pst_MmPdeBXoLzjaP5E4ioc5EA

Problem query: "Exception in thread main java.lang.OutOfMemoryError" GATK HaplotypeCaller fails: how to fix

TL;DR: HaplotypeCaller ran out of Java heap, not disk or CPU. GATK uses a small default heap unless you pass a larger one, so raise the heap with the java options flag on the GATK command, then prove the fix on one small interval before rerunning the whole genome. If a modest heap increase still fails, split the job by interval instead of raising the heap without limit.