Hypertelorism, Severe, With Midface Prominence, Myopia, Mental Retardation, And Bone Fragility

Software versions used: SAM 3.5 (Jul 2005) [37] (link), NCBI BLAST+ 2.2.24+ (Aug 2010) [3] (link), FASTA 36.3.3 (Feb 2011) [40] , WU-BLAST 2.0MP-WashU (May 2006), HMMER 2.3.2 (Oct 2003), and HMMER 3.0 (Mar 2010).
Example sequence alignments and profile HMMs were sampled from Seed alignments and profiles in Pfam 24 [11] (link). Example target sequences were sampled from UniProt version 2011_03 [43] (link). One experiment that characterized roundoff error used older versions, Pfam 22 and UniProt 7.0.

In order to understand the modeling choices underlying our new imputation algorithm, it is crucial to consider the statistical issues that arise in imputation datasets. For simplicity, we will discuss these issues in the context of Scenario A, although we will also extend them to Scenario B in the Results section. Fundamentally, imputation is very similar to phasing, so it is no surprise that most imputation algorithms are based on population genetic models that were originally used in phasing methods. The most important distinction between phasing and imputation datasets is that the latter include large proportions of systematically missing genotypes.
Large amounts of missing data greatly increase the space of possible outcomes, and most phasing algorithms are not able to explore this space efficiently enough to be useful for inference in large studies. A standard way to overcome this problem with HMMs [6] (link),[11] (link) is to make the approximation that, conditional on the reference panel, each study individual's multilocus genotype is independent of the genotypes for the rest of the study sample. This transforms the inference problem into a separate imputation step for each study individual, with each step involving only a small proportion of missing data since the reference panel is assumed to be missing few, if any, genotypes.
In motivating our new imputation methodology, we pointed out that modeling the study individuals independently, rather than jointly, sacrifices phasing accuracy at typed SNPs; this led us to propose a hybrid approach that models the study haplotypes jointly at typed SNPs but independently at untyped SNPs. We made the latter choice partly to improve efficiency – it is fast to impute untyped alleles independently for different haplotypes, which allows us to use all of the information in large reference panels – but also because of the intuition that there is little to be gained from jointly modeling the study sample at untyped SNPs.
By contrast, the recently published BEAGLE [13] (link) imputation approach fits a full joint model to all individuals at all SNPs. To overcome the difficulties caused by the large space of possible genotype configurations, BEAGLE initializes its model using a few ad-hoc burn-in iterations in which genotype imputation is driven primarily by the reference panel. The intuition is that this burn-in period will help the model reach a plausible part of parameter space, which can be used as a starting point for fitting a full joint model.
This alternative modeling strategy raises the question of whether, and to what extent, it is advantageous to model the study sample jointly at untyped SNPs. One argument [20] (link) holds that there is no point in jointly modeling such SNPs because all of the linkage disequilibrium information needed to impute them is contained in the reference panel. A counterargument is that, as with any statistical missing data problem, the “correct” inference approach is to create a joint model of all observed and missing data. We have found that a full joint model may indeed improve accuracy on small, contrived imputation datasets (data not shown), and this leads us to believe that joint modeling could theoretically increase accuracy in more realistic datasets.
However, a more salient question is whether there is any useful information to be gained from jointly modeling untyped SNPs, and whether this information can be obtained with a reasonable amount of computational effort. Most imputation methods, including our new algorithm, implicitly assume that such information is not worth pursuing, whereas BEAGLE assumes that it is. We explore this question further in the sections that follow.

Repeats in the genome assembly of P. sorghi were defined with RepeatModeler v1.73 (Smit and Hubley 2008 ) and masked with RepeatMasker v4.0.9 (Smit et al. 2013 ). The same library was used to identify repeats in the transcriptome assembly. Gene models were annotated in the genome assembly using MAKER (Cantarel et al. 2008 (link)), with additional putative effectors identified using hidden Markov models (HMM) with HMMER (Eddy 2011 (link))and regular expression string searches of ORFs (Fletcher and Michelmore 2018 ). The MAKER pipeline was provided with the RepeatModeler profile as well as assembled transcripts and translated ORFs from the transcriptome of P. sorghi, all described above, plus ESTs (option: altest) and protein sequences of other oomycete species available from NCBI. MAKER was initially run without a SNAP HMM, inferring genes using est2genome and protein2genome. These predictions were used to train a SNAP HMM (Korf 2004 (link)) that was used for a subsequent run of MAKER with both est2genome and protein2genome set to 0. The predicted proteins were again used to train a new SNAP HMM (Campbell et al. 2014 (link)). This process was repeated twice to generate three SNAP HMMs, which were used sequentially in three independent runs of MAKER. The annotations produced were evaluated as previously described (Fletcher et al. 2018 (link)) to select a single optimal run. This involved contrasting the number of gene models predicted, mean protein length, BLASTp hits to other oomycete annotations, and Pfam domains annotated by InterProScan (Finn et al. 2014 (link); Jones et al. 2014 (link)). Annotation of genes encoding putative effectors was performed as previously described (Fletcher et al. 2018 (link)). Briefly, the entire genome was translated into ORFs. These ORFs were surveyed for secretion signals using SignalP3.1 and SignalP4.0, and crinkler (CRN) motifs of LWY domains using HMMs. For peptides with secretion signals, the 60 residues beyond the predicted cleavage site were surveyed for an RXLR or RXLR-like motif and subsequently for a downstream EER or EER-like motif. ORFs encoding peptides that were predicted to be secreted and contained an (L)WY domain or a CRN motif were considered high-confidence putative effectors (HCPEs). ORFs encoding peptides that were predicted to be secreted and encoded an RXLR and EER domain, but did contain an (L)WY domain, or encoding peptides not predicted to be secreted, but contained an (L)WY domain, or a CRN motif were considered low-confidence putative effectors (LCPEs). The putative effectors and MAKER annotations were reconciled so that annotations did not overlap on the same strand. This was performed so that (1) any HQE or LQE annotations that did not overlap a MAKER annotation were added to the master annotation; for P. sorghi this was 12 HCPEs and 122 LCPEs. (2) HCPEs that overlapped MAKER annotations with the same start coordinates but earlier stop coordinates were discarded; for P. sorghi this was six peptides. (3) HCPEs that overlapped MAKER annotations with the same start coordinates but later stop coordinates replaced the model proposed by MAKER if they had a higher BLASTp score to the NCBI nr database than the overlapping MAKER model; for P. sorghi this was six peptides. (4) HCPEs that overlapped MAKER annotations but had different start coordinates and later or identical stop coordinates were retained over proposed MAKER models; for P. sorghi this was 27 peptides. (5) HCPEs that overlapped MAKER annotations but had different start coordinates and earlier stop coordinates were investigated to determine if the MAKER model should have a modified start coordinate; for P. sorghi this was six peptides. (6) Any LCPEs that overlapped MAKER annotations were discarded; for P. sorghi this was 142 ORFs. The same effector prediction workflow was then applied to the reconciled annotation set to determine the reported effector counts. Tracks for repeats, transcript coverage, annotation, and effector annotations were generated in 100 kb windows along each chromosome using Bedtools v2.29.2 and plotted using Circos (Krzywinski et al. 2009 (link)).

The genetic context analysis was done in the host genomes of the latent genes in the core-resistomes. The analysis included 1429 unique resistance gene sequences corresponding to 136 ARGs. For each ARG, the closest known homolog was identified using BLASTx v2.10.1 [36 ] to align the gene sequences against the CARD database [26 ]. Here, CARD was used since it is more comprehensive than ResFinder and includes, in contrast to ResFinder, some genes that are not clinically relevant and/or mobile. Then, genetic regions of up to 10,000 base pairs upstream and downstream of the gene sequences were retrieved using GEnView v0.1.1 [42 (link)] and screened for the presence of genes associated with MGEs and integrons. The genetic regions were translated in all six reading frames using EMBOSS Transeq v6.5.7.0 [43 (link)] and searched with 124 HMMs from MacSyfinder Conjscan v2.0 representing genes involved in conjugation [44 (link)], using HMMER v3.1b2 [45 (link)]. Insertion sequences (ISs) and other mobile ARGs were identified by applying BLASTx v2.10.1 [36 ]. For IS elements, a reference database based on ISFinder [37 , 38 ] was used to find the best among overlapping hits, with the alignment criteria that hits should display $>50\%$ coverage and $>90\%$ amino acid identity to a known IS transposase, as well as being located within 1,000 base pairs of the latent resistance gene (upstream or downstream). For co-localized mobile ARGs, ResFinder v4.0 was used as a reference database [25 ], with the alignment criterion that hits should display an amino acid identity $>90\%$ to a known ARG. Finally, the genetic regions were searched for integrons using Integron Finder v1.5.1 [46 (link)]. After the screening, the genetic contexts were manually investigated and curated.

Protein Homology/analogY Recognition Engine V 2.0 (Phyre2) tool was used for homology modeling of concerned PTB protein [43 (link)]. Following two attributes in protein selection for homology modeling were also kept considered for inclusion:
After following the above-mentioned inclusion criteria, CNN1 was selected for 3D homology modeling. The target sequence of Calponin1 (CNN1 Isoform 2) protein was searched from UniProt database (Entry No. P51911-2). PHYRE2 tool was used for homology modeling that consisted of sub-algorithmic stages:
Stage 1—Gathering Homologous Sequences: CNN1 isoform 2 sequence was scanned against the specially curated NR20 (No of sequences with >20% mutual sequence identity) protein sequence database with HHblits [44 ]. The resulting Multiple Sequence Alignment (MSA) was used to predict the secondary structure with PSI-blast based secondary structure PREDiction (PSIPRED) [45 (link)], and both the alignment and secondary structure prediction combined into a query Hidden Markov Model (HMM).
Stage 2—Fold Library Scanning: The models were scanned against a database of HMMs [46 ] of proteins of known structure. The top-scoring alignments from this search were used to construct crude backbone-only models.
Stage 3—Loop Modeling: Indels in these models were corrected by loop modeling.
Stage 4—Side-chain Placement: Amino acid side chains were added to generate the final PHYRE2 models.
The best homology models were selected based on the top-ranked modeled structures (according to similar template pattern). The alignment description of templates obtained through manual protein-blast comparison with the highest percent identity at 100% confidence was generated by Phyre2 tool. The quality assessment of stereochemical properties of 3D homology models were carried out by PROCHECK [47 ]. Ramachandran plot and residual properties of constructed 3D model along with the dihedral angles of φ against ψ of all possible conformations of amino acids in protein structure had also been studied in Ramachandran plot [48 ]. The validation of 3D protein models were also performed by SAVES (Structure Validation Server: https://saves.mbi.ucla.edu/) web server [49 ] to know the probable structural errors and z-score of the chosen model. SMART-EMBL tool (http://smart.embl-heidelberg.de/) was used to compute the confidently predicted domains, repeats, motifs and low complexity region (LCR) in protein. The SuperPose webserver (http://superpose.wishartlab.com/) was used to calculate both sequence alignment between template and 3D homology model through structure superposition using modified quaternion eigenvalue approach to generate RMSD statistics of the superimposed molecules [50 (link)].