Two new Max Planck tools target the 60% to 70% of sequenced patients left undiagnosed

Roughly 30% to 40% of patients get a clear molecular diagnosis, according to the phys.org account of the two papers ^[3]. Subtract, and 60% to 70% are left undiagnosed ^[4]. That account attributes part of the gap to a long focus on point mutations, the substitution of single letters in the sequence ^[5]. Structural variants work at a different scale: whole sections of the genome missing, duplicated or displaced ^[6]. How much of the unsolved fraction either tool converts into a diagnosis depends on how many of those cases are caused by structural variants or tandem repeats to begin with.

dicast scores candidate calls. "The core idea is that this method allows us to learn the patterns behind real structural variants and thus correctly classify new variants," Alavi said ^[9]. That aims at precision, since earlier methods produced false positives that someone then had to check by hand ^[10]. It leaves the detection ceiling where it was, because a variant that no caller proposes is never scored, and Alavi named that ceiling himself. "Because structural variants are often larger than the read segments themselves, they are difficult to detect in this data," he said of short-read sequencing, which he called the most common approach in genetic diagnostics ^[8][7].

In patient data, dicast detected all the pathogenic structural variants and filtered out a large number of false-positive artifacts ^[11]. The report leaves out the number of patients, the number of calls kept, and the caller dicast was measured against ^[12]. Sensitivity on variants already known to be pathogenic is the more tractable half of the problem. Because the manual check is what costs a curator time ^[10], the figure a lab will want is candidates retained per case.

The second tool counts. "In the second paper, we have developed an algorithm that can quickly and precisely count, based on sequencing data, how often a basic motif is repeated. We also provide a tool that visualizes this data," Vingron said ^[16]. Repeat numbers differ greatly among healthy individuals and in disease states ^[13], and replication errors in these tracts cause various diseases, especially neurological ones ^[15]. The same counting underlies paternity testing and forensics ^[14]. Al Raei and Ghareghani demonstrated the identification of pathological repeats using concrete examples ^[22].

Of the two, the repeat counter makes the more checkable claim, because a repeat genotype is a number a lab can hold against a validated reference. "Current sequencing technologies, combined with our specialized analysis algorithms, promise to further improve genetic diagnostics," Vingron said ^[21].