只要内存足够大，IO足够快，序列比对照样可以跑很快

序列比对是NGS数据分析中比较耗时，复杂度较高的一步，李恒所写的bwa mem已普遍用于各种临床应用中的序列比对一步，但性能应用于临床检测还需进一步提升。

Intel曾经用GATK best practice对每一步所耗时间做了精确的测试，https://wcm-ciqa.intel.com/content/dam/www/public/us/en/documents/white-papers/deploying-gatk-best-practices-paper.pdf

推荐一篇今年发表于Bioinformatics上的工具（BWA-MEME），这个工具相比bwa-mem2（指令集加速版bwa mem），与 BWA-MEM2比对结果一致，但需要更大内存。

https://academic.oup.com/bioinformatics/advance-article-abstract/doi/10.1093/bioinformatics/btac137/6543607?redirectedFrom=fulltext

github链接地址为：

https://github.com/kaist-ina/BWA-MEME

以下feature截取自github：

BWA-MEME: BWA-MEM emulated with a machine learning approach

• BWA-MEME produces identical results as BWA-MEM2 and achieves 1.4x higher alignment throughput.
• Seeding throughput of BWA-MEME is up to 3.32x higher than BWA-MEM2.
• BWA-MEME builds upon BWA-MEM2 and includes performance improvements to the seeding.
• BWA-MEME leverages learned index in suffix array search.
• BWA-MEME also provides feature to accomodate various memory size in servers.

与 BWA-MEM2一样的结果，可惜就是内存需求太大了，最小都要38G起步，以下是github release截图

不差钱的用户可买大内存机试试，据说内存比前两年还是便宜一点点了