Unlike most other analytics in medicine, where automation is king even in the interpretation of the results, genetic testing still remains complex under all aspects. In particular for what concerns the interpretation of the results, despite the solid efforts made in developing artificial intelligence, the human factor is still pivotal and a good final outcome still operator-dependent.
We can divide the process of validating the genetic results in three phases: 1. technical check of the analysis; 2. clinical interpretation; 3. report editing and validation.
- Technical check of the analysis
A bad technical execution of the test is not necessarily appearing as a bad sequencing graph, odd result or inconsistent data. Sometimes, even when the raw data looks OK, there may be a gross mistake that remains unrecognized for a long time just because “everything looks OK”. This is the reason why it is very important that even clinical scientists and medical geneticists who do not participate in the material execution of the test have a sound background in laboratory techniques, so that they are ready to suspect the most common (and uncommon) technical pitfalls.
Genetic testing is a delicate procedure which take shape in a multi-step manner and it’s therefore pivotal to be aware that human mistakes and machine malfunctions may occur at any stage, from DNA extraction to DNA amplification (PCR) and sequencing (Sanger, NGS). The most common technical failures are caused by the use of degraded reagents, worn consumables, contaminated devices (or environments), partial and unseen machine breakdown (like, for instance, insufficient or inhomogeneous heat-retention in thermo-cyclers) or bad instrument/software calibration. Mistakes such as sample exchange or tube mislabelling are not very frequent, although they can never be ruled out as a possible cause of bad or inconsistent results.
The technical check must also include the testing method. It happens, and not so infrequently, that a sample is processed properly but using the wrong method (e.g. that sequencing is performed instead of fragment length analysis). This can easily happen, for instance, with MLPAs. For some genes, different MLPA kits exist to cover different exons or to include also methylation testing. If the kit has been ordered or used without the counselling of the geneticist, false negatives may be easily diagnosed. In some other cases the method is just partially inappropriate, i.e. it is indicated to detect some but not type of mutations of a certain gene, leading again to a high rate of false negatives. For example, it is well known that certain repeat expansion causing severe neurological diseases may be detectable be sequencing up to a certain size, whereas the expansion of several thousand repeats requires the use of alternative methods like Southern blotting. To avoid the adoption of the wrong testing method it is of help that any incoming request of testing is pre-checked and approved by the medical geneticist before technicians initiate the sample processing.
- Clinical interpretation of the results
Unfortunately there are no unique rules that apply for the interpretation of each kind of mutation. The genetic results must always be interpreted in the context of the clinical information (including the most likely mode of inheritance), the gene(s) in question and the detection power of the method utilized.
The clinical information of the patient is essential, not only to assign a significance to new mutations not previously reported in the literature, but also to revise the classification of known mutations that have been thus far wrongly considered as disease-causing. Especially when doing a “whole” approach (whole-exome or whole-genome sequencing) the evaluation of the most likely mode of inheritance is extremely helpful in filtering variants which are most likely to be disease-causing (i.e.: if the patient is expected to be affected more by an autosomal recessive disorder, it is less likely that a single heterozygous mutation is the causative one).
It is also crucial to keep in mind that, despite most genes can harbour all kinds of pathogenic mutations, the mutational spectrum of some genes can be quite atypical. In some genes, for instance, only nonsense mutations are likely to be disease-causing, whereas missense mutations are usually neutral. In some other genes pathogenic mutations are found in just one exon or two, whereas in some other again point mutations are exceptional and large deletions or duplications are the rule.
In certain cases the effect of the mutation is strictly location-dependent. Nonsense and frame shift mutations are typically the easier to be interpreted, because they are usually disease-causing. However, especially when they fall by the end of the coding sequence, they can also be benign. It is also well known that even synonymous changes, which are benign in most cases, may be definitively pathogenic when the nucleotide change is activating a cryptic splice site (even today, especially in whole genome and whole exome sequencing, a percentage of these mutations may go lost, because splicing predictions can require a sophisticated approach and not rarely a manual computation for each and every single variant).
Hypomorphic alleles. There are also mutations which cannot be considered entirely pathogenic neither benign. This is the case of the so-called hypomorphic alleles (see for instance the PKD1 gene), which are generally consistent with missense mutations. These variants actually exert a modulatory effect when they are found in co-presence with a pathogenic mutation.
The paradigm of an uncertain result is given by the finding of a missense mutation which has not been previously described either in the literature or in the most common databases (dbSNP, ESP, HGMD, ExAC). In such a case definitive conclusions cannot be drawn until further investigations are carried out (e.g. family testing or RNA analysis).
There are some guidelines and some software to help geneticists in the difficult task of interpreting the results of the test. Practice guidelines have been released by several medical societies, from the ACMG to the ESHG, whereas a myriad of tools to perform bioinformatic analysis of the data and in silico predictions are available today to enrich all kinds of genetic testing, from single gene to whole genome sequencing.
- Report editing
The last effort of the medical geneticist is in summarizing the most relevant results and writing the conclusions. A good genetic report (at least of the kind that can be easily understood by the requesting physician and the patient – which is also the kind of report being most frequently requested on the market) contains a conclusion paragraph, where the clinical question about the patient (is he/she affected or not by this or that genetic disorder?) is answered in a clear and synthetic manner. If the clinical suspicion cannot be confirmed, the report should contain a reasoned explanation and recommendations on how to carry out further steps to reach conclusive diagnosis (e.g. further clinical examinations or family testing). When the results are unequivocally negative, hints about the differential diagnosis and extra genes to be tested may be given. Finally, the genetic report should be checked for the correctness of the patient’s data and the respect of the editing standards which are released from time to time by the working groups of human genetics societies around the world (for instance, about the use of the correct mutation nomenclature – www.genenames.org -, the use of a standardized language and the recommendation about genetic counselling).
Summarizing we can say that making a genetic report is still a question of high professionalism, deep care and, of course, experience in results validation. Despite the several efforts so far made to implement complex algorithms of artificial intelligence, the human touch remains indispensable.