Developing Del2Phen: a novel phenotype description tool for chromosome deletions

2.1. Del2Phen input and annotations

For this study, Del2Phen uses genotype and phenotype data stored in the Chromosome 6 database as input (Figure 1). This phenotype data is collected from Chromosome 6 Project participants (parent cohort) and eligible case reports (literature cohort) using the Chromosome 6 Questionnaire (see Supplementary Information S1 for further details). This multilingual web-based questionnaire contains 132 main questions that cover 654 phenotypic features, developmental characteristics and other types of clinical information (Supplementary Table S1). All phenotypes are coded using Human Phenotype Ontology⁸ (n = 423), ICD-10⁹ (n = 11), ORPHA¹⁰ (n = 1) or custom (n = 219) terms. Custom codes were developed to encode information for which no standard codes were available, such as growth parameters, age at onset of certain conditions, or developmental milestones. Genotype data from both cohorts derive from the results of microarray or other high-resolution genetic analyses, performed in different laboratories using various platforms. Genomic locations based on a reference genome other than hg19 are converted to hg19 using the UCSC LiftOver Tool (https://genome.ucsc.edu/cgi-bin/hgLiftOver).¹¹

Del2Phen annotates several aspects of each chromosome aberration, including type (deletion versus duplication), locus (base pair position at the start and end of the aberration in hg19) and genes intersecting the aberration. For each gene, the following haploinsufficiency (HI) scores are annotated: HI score from Huang et al. (2010)¹², probability of loss-of-function intolerance (pLI) score based on Lek et al. (2016),¹³ and predicted probability of haploinsufficiency (pHaplo) score from Collins et al. (2022).¹⁴ Each score denotes the probability of a gene having a clinical effect in case of deletions or loss-of-function variants. Genes with an HI score between 0-10%, a pLI score ≥0.9 or a pHaplo score ≥0.86 are most likely to have a phenotypic effect.^12–14 For Del2Phen, genes are considered HI-genes if they fulfil at least two of these three criteria when all three scores are known or at least one of two criteria when only two scores are known. Selected HI-genes known to have a prominent, highly penetrant effect on the phenotype are annotated as dominating-effect genes in the tool. Details on the selection of these genes and additional technical information are provided in Supplementary Information S1.

2.2. Workflow and genotypic similarity settings

Del2Phen performs pairwise comparisons between patients using specific parameters to score their genotypic similarity. For aberrations of the same type on the same chromosome, the Jaccard index is used to measure the size, location and (HI-)gene overlap between pairs (see example in Figure S1). For each of these parameters, minimum thresholds can be introduced. Dominating-effect gene content is the only binary parameter, indicating whether or not the sets of affected dominating- effect genes are equivalent (concordant) between patients. For patients harbouring multiple aberrations of the same type on the same chromosome, Del2Phen uses the total size and gene content of these aberrations for the comparison.

Given the prominent role that dominating-effect genes play in the clinical picture, Del2Phen’s first genotypic similarity parameter is concordance in the presence or absence of these genes. Only individuals with the same affected dominating-effect genes are identified as genotypically similar by the tool. Similarly, because of their predicted phenotypic effect, Del2Phen’s second genotypic similarity parameter is the fraction of HI-genes shared between the index and each individual concordant for the selected dominating-effect genes. This parameter, HI-gene similarity (Figure S1), can range from 0 to 100% depending on the extent of HI-gene overlap between deletions. Del2Phen uses these two parameters to identify individuals in the Chromosome 6 database as genotypically similar to an index, thereby generating a phenotype description tailored to that index patient (Figure 1).

2.3. Assessing optimal HI-gene similarity for chromosome 6 deletions

Ideally, Del2Phen would only generate a phenotype description based on individuals with 100% HI- gene overlap, ensuring the highest level of functional similarity. However, this would result in too few genotypically similar cases. To identify the thresholds that would allow as many individuals with a chromosome 6 deletion as possible to receive a phenotype description based on highly similar deletions, we evaluated the impact of various minimum HI-gene similarity thresholds on the: (i) size of the genotypically similar group (i.e. group size or number of connections) and (ii) precision of the resulting phenotype descriptions, quantified using positive predictive value (PPV). We used 115 main features to determine PPV (features in bold in Table S1) to prevent double-counting of the same feature if both the main and sub-features were known (see Supplementary Information S1 for additional criteria).

PPV was calculated individually for all patients in the database who had at least five connections for any given HI-gene similarity (each patient was used as an index once) using the following formula: PPV = number of features in the description that are also present in index / total number of features in the description, where ‘total number of features in the description’ represents the number of bold features in Table S1 that were present in ≥20% of the connected group and at least two individuals. The ‘at least two individuals’ criterion was added to exclude features present in only one patient in case of a minimum number of connections (n = 5). For these calculations, features not reported in literature cases were considered not present.

The objective of our assessments was to determine the extent to which we can decrease the minimum HI-gene similarity in order to reach higher numbers of genotypically similar cases while still preserving PPV. Given the non-normal data distribution, we used the median (50^th quartile) as a measure of PPV. To ensure that the presence of a dominating-effect gene does not affect the overall results, this metric was also determined separately for individuals with and without such a gene in their deletion, and any significant differences between these groups were evaluated using a Mann- Whitney U test. As the sample sizes in the literature and parent cohorts were insufficient for separate assessments, we examined differences in the number of phenotypic features (Mann- Whitney U test) and dominating-effect gene content (Chi-square test) between the two cohorts to address potential confounding. All statistical analyses were performed with Python 3.10¹⁵ using the SciPy,¹⁶ Matplotlib¹⁷ and Plotly packages.¹⁸

Lastly, to demonstrate Del2Phen’s application using the genotypic similarity parameters determined in this study, we used it to generate a phenotype description for two randomly selected patients in the Chromosome 6 database: one with a dominating-effect gene in their deletion and one without.

3.1. Participant and genotype characteristics

As of November 2023, the Chromosome 6 database included detailed genotype and phenotype information for 184 project participants who had completed the Chromosome 6 Questionnaire (parent cohort) and 342 individuals described in literature (literature cohort). From these, 452 individuals (142 from the parent cohort and 310 from the literature cohort) were eligible for use by Del2Phen (Table 1 and Figure S2). Their deletions were not equally distributed throughout the chromosome (Figure S3), with the proximal 6p region having a notably low number of cases. The number of phenotypic features per patient was significantly higher in the parent cohort compared to the literature cohort (p <.0001) (Table 1), but there was no significant difference between cohorts in the number of cases encompassing at least one dominating-effect gene (p =.28).

3.2. Assessment of genotypic similarity parameters

Of the 121 genes (Table S2) that fulfilled our definition of a HI-gene (Materials and Methods 2.1.), five were selected as dominating-effect genes and used as the tool’s first genotypic similarity parameter (Table S3).

As expected, the number of individuals who exceed a certain number of connections (5, 15, 20 or 30) decreases as we increase the minimum HI-gene similarity threshold (Table 2 and Figure S4). Conversely, the overall median PPV, calculated across all eligible patients, shows a positive trend with HI-gene overlap, peaking at 0.333 when minimum HI-gene similarity is set to 70% or higher (Table 2). Interestingly, the increase in median PPV with HI-gene similarity is only marginal in groups containing a dominating-effect gene (Figure 2A), with PPV being significantly higher in these groups at lower HI-gene similarity thresholds (≤50%) when compared to groups without such a gene (Table S4). This difference is no longer statistically significant when the minimum HI-gene similarity is at least 60%.

• Download figure
• Open in new tab

Fig. 2: Median PPV based on dominating-effect (DE) gene content (A) and minimum group size (B) for various minimum HI-gene similarity thresholds.

A. Median PPV for individuals harbouring a DE- gene in their deletion exhibits only slight variation in response to changes in the minimum HI-gene similarity threshold. In contrast, for individuals who do not lack such a gene, median PPV notably increases when the minimum HI-gene similarity threshold surpasses 50%. B. When comparing groups of different sizes, larger groups (≥15 or ≥20 individuals) tend to exhibit higher median PPVs, particularly at higher HI-gene similarities (≥80%). At these similarity thresholds, increasing the minimum group size from 20 to 30 does not affect PPV. For HI-gene similarities thresholds of ≥60% and ≥70%, groups of ≥15 and ≥20 individuals exhibit the same median PPV (0.333).

When evaluating groups based on different minimum sizes, the highest median PPV (0.4) occurs in individuals with ≥20 connections at HI-gene similarity thresholds of ≥90% and 100%, and increasing the group size to ≥30 does not further improve this metric (Table 2 and Figure 2B). When the minimum HI-gene similarity is <80%, the differences between groups of different sizes become less pronounced. At ≥60% and ≥70% HI-gene similarity, groups containing ≥15 individuals and those with ≥20 exhibit a similar median PPV (0.333) (Figure 2B), with a minimum group size of 15 individuals sufficient to achieve a statistically significant increase in PPV compared to smaller groups at these thresholds (p-values provided in Table S5).

3.3. Del2Phen’s default genotypic similarity settings for chromosome 6 deletions

Considering the results above, we determined the genotypic similarity algorithm depicted in Figure 3 to guide Del2Phen’s generation of phenotype descriptions for chromosome 6 deletions. Under the default settings, at HI-gene similarity ≥80%, Del2Phen will provide a clinical description with the highest achievable functional similarity for groups of ≥20 individuals. At 60% and 70% HI-gene similarity, the group-size cutoff will be reduced to 15. Below these similarity thresholds, phenotype descriptions will only be generated if a dominating-effect gene is present, coupled with the highest HI-gene similarity that would yield groups of at least 15. The rationale behind this algorithm is discussed in section 4.3.

• Download figure
• Open in new tab

Fig. 3: Genotypic similarity algorithm for Del2Phen’s generation of phenotype descriptions for chromosome 6 deletions.

The tool will leverage the maximum possible functional similarity within groups that exceed a certain size for generating a comprehensive phenotype description. At HI-gene similarity ≥80%, a description will be provided for groups of at least 20 individuals. The group size cutoff will be lowered to 15 for HI-gene similarities of at least 70% and 60%. Below these similarity thresholds, phenotype descriptions will only be given if a dominating-effect (DE) gene is present, coupled with the highest HI-gene similarity yielding groups of at least 15 individuals. In each step, n represents the number of additional individuals in the Chromosome 6 database who become eligible for a phenotype description as the algorithm progresses through the step, while N represents the total number of individuals who would or would not receive a phenotype description based on these settings and our current data pool.

Table 3 and Table S6 show the phenotype descriptions for two patients from the Chromosome 6 database generated by Del2Phen based on this algorithm. The first patient (individual Id226) has a 2.4 Mb 6q24.3q25.1 deletion that includes the dominating-effect gene TAB2 (TAK1- BINDING PROTEIN 2, HGNC:17075, MIM* 605101). Del2Phen identified 17 individuals in the Chromosome 6 database (2 Chromosome 6 Project participants and 15 literature cases) who shared the same dominating-effect gene and at least 70% of their HI-gene content with this patient. The phenotype description for this patient (Table 3) consists of 25 features present in at least 20% of the connected group. Of these, 12 features were also seen in the index, leading to a PPV of 0.48 based on the full description and a PPV of 0.6 when considering only the selected features used to determine the tool’s setting (Table S1).

The second phenotype description (Table S6) was generated for a patient with an 11.1 Mb deletion in the 6q13q14.1 region that does not span a dominating-effect gene. The description is based on the phenotypes of 16 individuals (5 project participants and 11 literature cases) with a minimum of 70% overlap in their HI-gene content. The clinical description for this patient was lengthier, encompassing 38 features, and resulted in a PPV of 0.45, rising to 0.5 when using only the selected features in Table S1.

The clinical description for this patient is based on the phenotype and other clinical features of 17 individuals that share the same content of dominating-effect genes and at least 70% of the affected HI-genes. The features in the description are present in at least 20% of the individuals connected to this patient (i.e. at least 4). The clinical description in this table is generated based on all features covered by the Chromosome 6 Questionnaire deriving from closed-ended questions (n=487). The features in bold were used to calculate PPV in this study. In italics are the features of the index that were also seen in at least one connected individual but did not meet the prevalence threshold for inclusion in the description (i.e. were seen in less than 20% of the connected group).

4.1. Phenotype descriptions for chromosome deletions

Defining the functional similarity of chromosome aberrations is complex because of our still incomplete understanding of the function of many genes and the clinical consequences in case of deletions. While websites aimed at professionals, such as DECIPHER,¹⁹ often allow users to set the functional similarity range, Del2Phen will eventually be incorporated into a website primarily for parents. It is therefore important to pre-set the genotypic similarity parameters that will be used to generate phenotype descriptions. For Del2Phen, we utilised current knowledge of the predicted HI- effect of genes and known gene-phenotype associations to define genotypic similarity.

Because of the dominating role of the genes in Table S3 in the clinical picture, we expect individuals harbouring identical sets of these genes to have similar phenotypes. The observed higher PPV in groups harbouring a dominating-effect gene even at otherwise low HI-gene similarity (Figure 2A and Table S4), confirms the significant role of these genes in the clinical outcome and supports our decision to use concordance in dominating-effect gene content as Del2Phen’s first genotypic similarity parameter. This is also illustrated by the example in Table 3, where all the features associated with a deletion of the dominating-effect gene TAB2 (Table S3) were included in the phenotype description, and most were present in the index.

The second parameter, the HI-gene overlap between deletions, would ideally be set to 100%. However, the limited number of individuals who have fully overlapping deletions means that such high similarity would often result in phenotype descriptions based on small patient groups, if generated at all. Given the role of individual variation in the resulting phenotype and the incomplete penetrance of most observed features,²⁰ it was necessary to lower the HI-gene similarity cutoff to enable a larger number of individuals to receive a phenotype description and increase the sample size on which these descriptions would be based. Concurrently, it was important to ensure that adjusting the HI-gene similarity would not compromise the reliability of Del2Phen’s outcome.

4.2. Evaluating phenotype descriptions

Assessing the accuracy of phenotype descriptions in the context of structural chromosome aberrations is inherently complex, as even identical deletions can result in an array of phenotypic manifestations.^21,22 To gain insights into the relevance of features in the descriptions, we used PPV as an outcome measure to describe the fraction of features in the clinical description that were in accordance with the phenotype of the index. Remarkably, PPV did not drop below 0.2 when setting the HI-gene similarity threshold to 0% (Figure 2). While this may seem counterintuitive, these thresholds only indicate a minimum level of similarity, e.g. at a threshold of 10%, groups are formed of individuals who share 10% or more of their deletion’s HI-gene content. Increasing the minimum threshold results in individuals with lower levels of similarity being excluded from the groups, only retaining those sharing a greater fraction of HI-genes. This, in turn, contributes to a higher description PPV as the minimum HI-gene similarity increases, reaching a median of 0.333 for functionally identical deletions (Figure 2).

A maximum median PPV of 0.333 may seem low, but we have to consider the low a priori chance of multiple features occurring together in a single patient (index) given the reduced penetrance of many of the observed features and the variable expressivity associated with these aberrations.^20,21 For instance, if a characteristic is present in 25% of the genotypically similar group, it will be included in the phenotype description under our current settings but may not be present in the index patient, thus decreasing PPV. We believe that it is correct to include these features in the description to ensure comprehensiveness and not miss important findings, even though this might lower the tool’s precision metrics. Furthermore, the decrease in group size at such high HI-gene similarity (Figure S4) can also influence PPV.

Another potential factor is the variability in phenotypes associated with deletions in certain chromosome regions. While deletions in certain areas are linked to highly distinct phenotypes,⁴ deletions in other regions produce a much more diverse clinical outcome,⁵ thereby affecting the overall PPV. However, as Del2Phen’s similarity settings will be applied uniformly across the entire chromosome, PPV was determined for deletions in all regions.

Lastly, our relatively modest PPV might also be influenced by the amount and diversity of phenotypic information available for our parent cohort. In contrast to literature reports, in which authors are often restricted in the amount and kind of information they present (e.g. presenting only clinically significant issues), data collected via the Chromosome 6 Questionnaire yields much more detailed information (Table 1).⁷ Since almost all groups consist of both literature cases and Chromosome 6 Project participants, one subset of the group will have fewer but very specific characteristics (literature cohort), while the rest will present with a much more diverse phenotype (parent cohort), thus lowering PPV. Unfortunately, it was not possible to conduct the analysis separately for the two cohorts. Nevertheless, we can conclude that the extensive information collected from our parent cohort makes the phenotype descriptions generated by Del2Phen more complete than what can be derived solely from literature reviews.

4.3. Determining Del2Phen’s genotypic similarity settings for chromosome 6 deletions

Following Del2Phen’s genotypic similarity algorithm (Figure 3), a phenotype description will be generated based on groups of ≥20 individuals sharing the highest attainable HI-gene fraction in the range of 80% to 100%. This minimum group-size threshold is justified by our observations that PPV is higher in groups of ≥20, with no notable improvements when increasing group size to ≥30 (Table 2 and Figure 2B). At the minimum HI-gene similarity thresholds 60% and 70%, the median PPV remained constant when lowering the group size to ≥15 individuals (Table 2), while still being significantly higher compared to groups <15 (Table S5). For this reason, the group-size cutoff was lowered to at least 15 at these similarity thresholds, allowing Del2Phen to provide phenotype descriptions for a larger fraction of individuals in the database, as well as future new patients (Figure S4).

If an insufficient number of patients in the database share ≥60% of their HI-genes, Del2Phen will still provide a description for deletions that include a dominating-effect gene, based on the highest possible HI-gene overlap that would yield groups of ≥15 (Figure 3). This setting is based on the expectation that the phenotype in these groups would be primarily determined by the deletion of the dominating-effect gene rather than by the HI-gene similarity, as supported by our findings in Figure 2A and Table S4. At HI-gene similarity ≥60%, the difference in PPV between groups with and without a dominating-effect gene is no longer significant, suggesting that this fraction of shared HI- genes is sufficient to produce effects similar in impact to that of a dominating-effect gene.

Based on these settings and our current data pool, 250 individuals (55% of our current cohort) will be eligible to receive an automatically generated phenotype description (Figure 3). As the Chromosome 6 database continues to grow, the number of individuals eligible for a phenotype prediction will also increase.

4.4. Strengths and limitations

Del2Phen is an innovative phenotype description tool, developed primarily for delivering clinical information to parents. Unlike existing online tools, which are tailored to professionals and require a certain level of expertise, Del2Phen will deliver information via a parent-friendly website. Naturally, it can also be a valuable resource for healthcare professionals in counselling patients and their families. The clinical descriptions provided will be detailed and thorough compared to the limited information in current databases or literature reports, because Del2Phen uses the extensive well-curated data collected via the Chromosome 6 Project (Table 1).⁷ Even though it is currently optimised for chromosome 6, it can also be applied to other chromosomes if given the correct input, and could thus benefit people affected by a wider spectrum of chromosome disorders.

Since Del2Phen will be periodically updated with the latest information on HI or the clinical effect of genes, the clinical information it produces will be more up-to-date than that in scientific articles, which reflect knowledge at the time of writing. Although Del2Phen does not specifically account for non-coding features, which are often positioned on fixed locations on the genome, their effects will be incorporated once more evidence is available. Lastly, this tool can also be applied in research settings to expedite data analysis.

One limitation of Del2Phen pertains to the fact that its output can be sensitive to variability in resolution of the genotype data, which derive mostly from microarray analysis conducted using different platforms. We do not expect this to affect the outcome of this study, as only individuals for whom a high-resolution array was available were included in the Chromosome 6 Project.

Furthermore, an HI-gene located between the last absent and first present probes (or vice versa) was an exclusion criterion for the current study as it made the deletion of that gene unclear (Supplementary Information S1), thus ensuring that the grouping of genotypically similar individuals would not be affected.

Clinical data missing from literature reports might have an impact on the thoroughness of the phenotype descriptions. To avoid falsely increasing the prevalence of certain features, we considered features that were not mentioned in the literature reports as not present in the patients. We expect this assumption to hold true for highly clinically relevant features, but articles that focus on specific topics, like heart or brain defects, may sometimes omit unrelated phenotypes.

Finally, it is important to note that we must always be cautious when presenting clinical information on chromosome aberrations, whether derived from automatic or manual analysis. The unique nature of these aberrations, paired with the phenotypic variability, means that not all features will be seen in all patients. Such disclaimers will be added to our website, together with instructions for reading the clinical descriptions, links to websites with support information, and the option to download the medical information for discussion with healthcare professionals.

4.5. Future steps

The next step for Del2Phen is validating its output by comparing its outcomes to those from manual data analysis, the conventional approach for characterising the phenotypes of chromosome deletions. Following this, the tool will be deployed to describe the phenotypes of new, uncharacterised regions. In parallel, the development of the interactive website for parents will also commence. In a running study, with the help of involved parents and professionals, we are investigating the best way to report the clinical features on the website. The study’s findings will inform the website’s content and structure, including the minimum prevalence of clinical features in individuals with a similar aberration required for inclusion in the description. Lastly, while Del2Phen currently focuses solely on deletions, it will be modified for use on duplications once enough data on these aberrations is collected.