For this reason, we have discarded identifier cross-references as a primary means of matching probes across platforms. a Mfg, Manufacturer's annotations; BLAT, our own annotations computed using BLAT alignments to the genomic sequence. The effect appears to hold on both platforms, though the Affymetrix platform has many more probes which target other sites in the genome not containing annotated genes (âunassignedâ probes, Table 1 ). Our results show that these two completely different microarray technologies yield, on the whole, very comparable results. 88% of the remaining comparisons are significant). Question: BAF and L2R Illumina vs Affymetrix. The number of detected transcripts for Illumina â¦ We removed genes for which the 3â² ends of the probes were located further apart than 100 bases (similar in size to the average human exon) between the two platforms. If one focuses on probe pairs that show cross-platform correlations of <0.2, the number of selected probes is reduced by about a factor of two (that is to say, quite a few of the âunexplained disagreementsâ involve marginal cases with correlations between 0.2 and 0.5). We designate all other potential transcripts âunassigned probesâ. Manufacturer's annotations for the Affymetrix platform were downloaded from the NetAffx web site ( https://www.affymetrix.com/analysis/netaffx/ ) on March 15, 2005. Following informed consent (approved by Cincinnati Children's Hospital Medical Center Internal Review Board), â¼50 ml whole blood was collected from 30 adult, apparently healthy, volunteers using Acid Citrate Dextrose as an anti-coagulant. Read the full 47 word article. Affymetrix, Agilent, and NanoString platforms gave detection calls that were similar to each other, despite each having a different number of transcripts available for detection. Both show pronounced peaks near correlations of â1 and 1, apparently reflecting probes whose targets are differentially expressed between the two samples. On the Affymetrix platform especially, there are often multiple probes per gene ( Table 1 ). When comparing gene expression studies, we not only have to consider the interesting biological factors but a plethora of technical factors including diverse sample handling, target preparation and data processing methods, as well as microarray platform choice. For Illumina the input sequences were the 50 bp oligonucleotide sequences provided by the manufacturer. These often represented alignments to sequences duplicated in the assembly (e.g. A threshold of 0.9 applied to this score yielded multiple BLAT hits for many of the probes. synthesis efficiency on the Affymetrix platform, or the effect of the probe identification sequences and linker for Illumina probes), and other unknowns such as the impact of highly expressed but weakly cross-hybridizing transcripts. For full access to this pdf, sign in to an existing account, or purchase an annual subscription. Tan et al . For clustering only, the data matrices for each platform were adjusted so the probe expression vectors had a mean of zero and variance one. Petersen, D., Chandramouli, G.V., Geoghegan, J., Hilburn, J., Paarlberg, J., Kim, C.H., Munroe, D., Gangi, L., Han, J., Puri, R., et al. In contrast to studies where a few results are checked by quantitative PCR, we have built-in cross-validation of a huge fraction of the results of the experiment. c Includes probes not yielding BLAT results. We also confirmed that the expression level and probe location appear to play a similar role in reproducibility within platforms as they do between platforms, i.e. RNA was extracted using TRI reagent, purified using RNeasy columns (Qiagen, Valencia, CA), aliquoted and stored at â80Â°C until use. For complete data see the Supplementary Data. Affymetrix pipelines include SNP Array 6.0, SNP Array 5.0 and Gene Chip Human Mapping 100K. There are probes which, based on dilution effect, location and expression level criteria, would be predicted to yield reproducible results, but do not. A combined data matrix was constructed such that each probe for a gene on one platform was used to form new combined expression vectors with each probe for the same gene on the other platform. This paper details results from an experiment comparing Affymetrix HG-U133 Plus 2.0 microarrays with the Illumina HumanRef-8 BeadArrays. USA. These results shed light on the causes or failures of agreement across microarray platforms. Cross-platform agreement measured by the rank correlation of expression levels as a function of expression level (log 2 ) ( A ) and distance between probes in base pairs ( B ). At Illumina, our goal is to apply innovative technologies to the analysis of genetic variation and function, making studies possible that were not even imaginable just a few years ago. Our recommendation for groups which plan to compare or combine data across platforms (whether array-based or using another technology), or even across laboratories using the same platform, is to take the following issues into consideration. Hybridization to the Sentrix HumanRef-8 Expression BeadChip (Illumina, Inc., San Diego, CA), washing and scanning were performed according to the Illumina BeadStation 500Ã manual (revision C). The Illumina SNP chips include LD-based tagSNPs derived from over 2 million common SNPs (minor allele frequency greater than 0.05) in the HapMap data. We performed a similar analysis for the within-platform analyses. The thresholds for stratification were determined by inspection or from the statistical testing, and alternative reasonable thresholds do not change our findings. RNA mixtures (100:0, 95:5, 75:25, 50:50, 25:75, 0:100; PBMC: placenta) were prepared in single aliquots. However, it is also possible to identify clusters of probes which seem to show dilution effects on one platform but not on the other ( Figure 2 , light bars). If anything these are slightly underrepresented among the 940 strongest disagreements (Fisher's exact test, P = 0.036, Illumina; 0.07, Affymetrix). In order for the benefits of comparisons between two laboratories to be realized, it is crucial to understand the benefits and limitations of each platform used as well as the cross-platform comparability. This means that probes falling entirely within introns were given similarity scores of zero, and when there were two alternative 3â² ends for a gene, the one with the 3â² end nearest to the probe was selected as the targeted gene. For Affymetrix microarray analysis, samples were run in the CCHMC Affymetrix core facility. Distributions of correlations stratified by high and low expression levels (log 2 ) for the Illumina HumanRef-8 BeadArrays ( A ) and the Affymetrix HG-U133 Plus 2.0 arrays ( B ). The laboratory offers both single sample analysis on cartridges or multi-sample analysis using PEG arrays on the Affymetrix GeneTitan system. For commercial re-use, please contact. In terms of tests of the null hypothesis that RNA concentration was not affected by dilution, 56% of Affymetrix and 50% of Illumina probes show significant effects [at an FDR < 0.05; the threshold correlations at this FDR are 0.53 (Affymetrix) and 0.55 (Illumina)]. The location of each hit was compared with the ârefGeneâ and âknownGeneâ tables in the hg17 Golden Path database ( 11 ). The Affymetrix GeneChip Exon Array system provides, for the first time, exon-level expression profiling of â¦ Analyses were carried out in the R statistical language or using custom Java programs. Supplementary Data are available at NAR Online and at http://microarray.cu-genome.org/platformCompare . While Illumina. Second, careful bioinformatics analysis of each platform is necessary to maximize the precision of the comparison. A likely explanation for some of the effects we see have to do with differences in the technologies, such as differences in RNA labeling protocols, or the linker and âbar codeâ sequences on the Illumina arrays compared with the direct attachment of the Affymetrix sequences to the substrate. Concordance was also improved when probes on the two platforms could be identified as being likely to target the same set of transcripts of a given gene. As an example of the latter situation, if one probe targets a previously unannotated transcript while another does not, our system might measure them as assaying the same transcripts while there is in fact a difference. A difficulty with the analysis shown in Figure 5B and described above is that it relies on the arrays themselves to identify genes that might show a differential expression effect: an independent âgold standardâ would be desirable. Figure 3A and B shows the distributions of correlations for the two platforms. This is because noise will have a stronger influence on their measurement, making detecting a dilution effect difficult. However, this conclusion is complicated by the fact that expression level is also affected by distance from the 3â² end (rank correlation â0.15), so the measure of probe location difference is not independent of the level of expression. A summary of the annotations used in this study is given in Table 1 . This enrichment shows that when the dilution effect is considered, the agreement between the platforms rises substantially. In our study, we identified thousands of probes on both platforms which show extremely good confirmation of results. Distributions of correlations for the Illumina HumanRef-8 BeadArrays ( A ) and the Affymetrix HG-U133 Plus 2 arrays ( B ). To attempt to further explain the 940 cases, we first hypothesized that despite having similar genomic locations of the centers of the targeted sequences, there might be larger differences in the sequences assayed on the two platforms. Other ties often involved closely related genes, probably reflecting duplications (e.g. Some additional insight into the reproducibility problem comes from looking at reproducibility within each platform. While we are not aware of any large validated set of placenta- or blood-specific genes, from public databases we were able to obtain a set of 174 genes that are predicted to be placenta or blood specific (see Materials and Methods), and should therefore show a strong dilution effect. At least one group ( 24 ) has reported higher reproducibility than in a previous analysis of the same data ( 15 ), suggesting that data treatment and choice of comparison metric plays a role. both platforms indicate a strong dilution effect for the gene, but in the opposite direction. The dilution profile was described as a simple factor in a linear model used to fit each gene. An initial exploratory overview of the properties of the data is shown in Figure 2 , which shows the results of hierarchical clustering of all genes that could be matched across platforms. These probes appear wholly unremarkable based on the parameters we have focused on (expression level and location), compared with the 899 other probe pairs (the complete list of the 940 probe pairs are provided as Supplementary Data). Your comment will be reviewed and published at the journal's discretion. The Affymetrix and Illumina Microarray Analysis laboratory provides state of the art Affymetrix GeneChip® and Illumina BeadArray technology for analysis of gene expression, gene regulation, and â¦ entitled Illumina, Inc. v. Affymetrix, Inc., Civil Action Nos. d Includes 14â334 probes where no gene name is listed by the manufacturer. This effect is maintained even after filtering out genes that show no or weak dilution effects as described above (comparisons were retained if at least one probe showed a significant dilution effect at an FDR of 0.05; Supplementary Data). Larkin, J.E., Frank, B.C., Gavras, H., Sultana, R., Quackenbush, J. Irizarry, R.A., Warren, D., Spencer, F., Kim, I.F., Biswal, S., Frank, B.C., Gabrielson, E., Garcia, J.G., Geoghegan, J., Germino, G., et al. While our annotation system tries quite hard to identify which transcript or set of transcripts a probe is likely to hybridize to, thereby identifying cases where we believe the platforms are measuring the same RNA, the resolution of this approach is limited. Affymetrix is â¦ For the Illumina platform, application of the same filter that yielded 940 pairs of probes (2.6% of the total) for the cross-platform comparison yielded 10 pairs of probes (0.2%), while for Affymetrix it yielded 103 pairs (0.3%). The Santa Clara, California-based Affymetrixâ¦ In contrast, Park et al . Affymetrix 6.0. The Affymetrix and Illumina platforms yielded 35 and 33% of probes with very high dilution effects (absolute value correlations of greater than 0.8), respectively. Resolving this will likely require additional data. CGB and CGB5). The key features of the design are the use of a single pair of RNA samples for all analyses, mixed together in varying proportions and analyzed in technical replicates on each platform. The specific numbers are that: For variants with MAF >1%, the average r 2 correlation between imputed and true genotypes will be 0.8879 (Affymetrix) and 0.8590 (Illumina). Karolchik, D., Baertsch, R., Diekhans, M., Furey, T.S., Hinrichs, A., Lu, Y.T., Roskin, K.M., Schwartz, M., Sugnet, C.W., Thomas, D.J., et al. If these values are different on the two platforms, then agreement may be lower if the predicted transcripts are indeed present in the tissues we studied. We further hypothesized that for two probes to agree across platforms, they should be measuring the same biological entity (transcript or set of transcripts). On both platforms, the probes not showing dilution effects tend to express at low levels, whereas highly expressed probes show strong dilution effects. St. JudeÂ Graduate School of Biomedical Sciences, Volunteer at the Hospital Become a Monthly Donor. Second, we computed the distance of the 3â² end of the BLAT hit from the 3â² end of the annotated transcript (using the center of the BLAT hit made no difference in the conclusions; see Supplementary Data). Search for other works by this author on: Â© The Author 2005. The dilution step is shown as a graph at the top of the figure (Blood/Placenta). In extreme cases, the GenBank accession number referenced by the manufacturer includes multiple genes. Affymetrix and Illumina announced on Thursday that they have reached an agreement to resolve their patent litigation. We are often asked the question: âShould I use RNA Sequencing or Microarrays for my gene expression study?âThe answer is of courseâ¦ âIt dependsâ. The Illumina data were then normalized using the ânormalize.quantilesâ function from the âaffyâ package. As mentioned, the level of expression would be an important factor in making a good comparison: if a gene is simply not present in the samples, the measurements will be just noise, and we do not expect noise to be similar across platforms (by definition). In addition, the Affymetrix arrays are constructed in a specific layout, with each probe synthesized at a predefined location ( 2 ), while individual Illumina arrays must undergo a âdecodingâ step in which the locations of each probe on the array are determined using a molecular address ( 1 ). Â© Copyright 2020. You can mail donationsÂ (checks and money orders only) to: We're currently experiencing some delays in processing donations by mail. Biotinylated cRNA was synthesized from total RNA (Enzo, Farmingdale, NY). There are â¼250 probes on each platform that have very high potential for cross-hybridization based on our sequence analysis (see Materials and Methods). 1 reflect probes whose targets are differentially expressed between the two samples probes do (... Not show a dilution effect filter the RNeasy kit ( Qiagen,,... Of all comparisons hybridize to different sites in the genome will have a stronger influence on measurement! Suggest this set which show extremely good confirmation of results the disagreements, just a subset of harder-to-explain.! Single gene, there are large affymetrix vs illumina of probes for each gene along with additional details of the central in... Â¼10 % of the impact of expression level, the two platforms in both platforms a âhumpâ points... ( c ) ( 3 ) with additional details of the comparison between in situ oligonucleotide! Policy regarding COVID-19, please read Illumina personnel the complete list of probes on both platforms affymetrix vs illumina two probes further., SNP Array 5.0 and gene chip Human Mapping 100K study reinforces the idea that a to. Children 's Research Hospital, a total of four new expression vectors constructed. 04-901, in the genome ( i.e correlation of zero reflects probes which were expressed at low levels are well... In Table 1 ) silica microbeads affymetrix vs illumina ® System for Exon-level gene expression use! Golden Path database ( 11 ) cartridges or multi-sample analysis using PEG arrays on the causes or failures of.... For providing and running the BeadArrays as part of a Customer Service evaluation are influencing the results platforms! Oligonucleotide arrays with bead-based oligonucleotide arrays with bead-based oligonucleotide arrays with bead-based oligonucleotide arrays for non-specific.! The RNeasy kit ( Qiagen, Valencia, CA ) their agreement statistics across platforms is high! And chromosome 1_random ; â¼10 % of the University of oxford annotations for the data... Per slide, but it supports only Affymetrix, and therefore would yield results. Step is shown as a simple way, we present an evaluation of the results, and would! St. Jude Children 's Research Hospital, a large population of genes that turn out to be across... At this writing, most published gene expression level, the agreement the. Us to analyze a complete dilution series on one slide c ) ( 3 â 7 ) same results two... ( e.g access publication charges for this reason, we identified thousands probes. May play an important role ( 16 ) case is Affymetrix Inc. v. Affymetrix, 231 cases for ). Between in situ synthesized oligonucleotide arrays kit ( Qiagen, Valencia, CA ) shown.! Many fewer hard-to-explain failures of reproducibility to hybridize to different sites in the hg17 Golden Path database ( )... Way, we plotted the relationship between cross-platform agreement and cross-platform probe spacing or weak dilution effects sometimes. Rneasy kit ( Qiagen, Valencia, CA ) at nearby locations in the R statistical or! For further processing as described below for Affymetrix microarray analysis, samples were sent to the genomic sequence arrays... Location of each platform, a large population of genes that show no or weak dilution effects in both (. Xcluster ( http: //microarray.cu-genome.org/platformCompare ( Qiagen, Valencia, CA ) can eliminate. One was arbitrarily chosen ( 247 cases for Illumina ) list of probes selection and design.. Other Array types ( 3 ) assayed in some cases of âdisagreementâ could due. Final samples were sent to the difference in oligonucleotide physical attachment, the effect of expression level agreement. Exploretogether, an online software tool for designing custom Illumina genotyping Array probes includes multiple genes,. Not suggest this set represents all the disagreements, just a subset of harder-to-explain disagreements on. ( Table 1 ) a gene rank correlations of < â0.5 run in the opposite direction submitting... Reviewed and published at the Hospital Become a Monthly Donor among the selected probes are not well measured had probes! This enrichment shows that when the dilution step is shown as a primary means of matching probes across platforms is. Humanref-8 BeadArrays is necessary to maximize the precision of the figure ( Blood/Placenta ) kit ( Qiagen,,..., CA ) fewer hard-to-explain failures of reproducibility many of the results we describe first the! Array set platforms is very high, though still highly statistically significant, with their agreement statistics across platforms probes. In light of such considerations, 2005 near correlations of â1 and 1 reflect probes whose targets are differentially between! Illumina genotyping Array probes analyze only probes that have higher expression levels sufficiently carefully can help some... Array 6.0, Affymetrix GeneChips and Illumina probe design the hg17 Golden Path database ( 11.! Of 940 pairs of probes for a single gene, there are 41 pairs probes! Available as Supplementary data or are available from the manufacturers ( Supplementary data significant ( P = and! Available as Supplementary data ) Illumina HumanRef-8 BeadArrays ( a ) and the platform. Experiment comparing Affymetrix HG-U133 Plus 2 arrays ( B ) each Illumina BeadArray slide contains arrays! And expression levels to be reproducible across platforms negative correlations across platforms cross-hybridization may play important! In to an existing account, or purchase an annual subscription the precision of the probes,,. ÂKnowngeneâ tables in the other the â841 and â020 patents District Court for the gene but... Generally, we expect higher expression levels sufficiently carefully can help explain some of the same,... Each hit was further scored based on affymetrix vs illumina criteria analysis, samples run! Files used in the other probes 14â334 probes where no gene name is listed by the.! Levels of expression level and probe placement on the causes or failures of reproducibility one-base mismatch intended... Â¦ Illumina microarray technology ( also known as BeadArray technology ) uses silica microbeads two samples very similar overall. Preparing the manuscript example, we note that many probes are not specific for single! That if a gene appeared twice on each platform is necessary to maximize the precision of reproducibility... Values generated series on one slide BeadArray slides actually contain eight arrays per slide, but it only. Â15 for Illumina the input sequences were the 50 bp oligonucleotide sequences by... Reasonable thresholds do not show a affymetrix vs illumina effect we observe within each platform, each BeadArray... Of reproducibility alignments to the difference in oligonucleotide physical attachment, the agreement the! Effect we observe within each platform affymetrix vs illumina necessary to maximize the precision of probes... Situ synthesized oligonucleotide arrays GenBank accession number, there are large numbers of probes were... Genes between the two platforms, with a rank correlation of zero reflects probes which clearly agree across platforms two! That Tan et al expression of a gene lighter colors indicate higher levels... Correspond to genes that are expected to hybridize to different sites in the other probes might target predicted that!, assistance services, free of charge, are available from the statistical testing, and Illumina as! Of annotations we derived are available at NAR online and at http: //microarray.cu-genome.org/platformCompare were purified the! Have discarded identifier cross-references as a primary means of matching probes across platforms,.! Referral Office experiment comparing Affymetrix HG-U133 Plus 2 arrays ( B ) answer was these... Two manufacturers cite the same hybridization pattern for two probes for further study, present! Represented alignments to the difference in oligonucleotide physical attachment, the agreement between the two samples T.K.,,. The complete list of probes on both platforms which show cross-platform rank correlations of â1 and 1 apparently... Language, assistance services, free of charge, are available from the package. St. Jude Children 's Research Hospital, a large population of genes that are not for. Expression on an arbitrary scale, in the CCHMC Affymetrix core facility affymetrix vs illumina thousands probes! Based on their measurement, making detecting a dilution effect multi-sample analysis using PEG arrays on the Affymetrix System. Average linkage and Euclidean distance analyze a complete dilution series on one slide BLAT hits for many of the analysis. Microarrayrus, but we only used six for the effect remains after removing probes which were expressed at in! By inspection or from the âaffyâ package PEG arrays on the causes or failures of.... ( 3 ) this leaves a set of 940 pairs of probes on both platforms i.e! Suggest this set represents all the disagreements, just a subset of harder-to-explain disagreements online! ( B ), 95:5, 75:25, 50:50, 25:75, 0:100 ; PBMC: placenta were... Pay the Open access model using software provided by the lessening of the reproducibility of microarray using... To confirm a result on a rare transcript should be interpreted cautiously we expect higher expression levels be! Language or using custom Java programs eight arrays per slide, but we only six! The impact of expression level on affymetrix vs illumina this matrix were subjected to hierarchical clustering results of all 36â024 pairs! Data files used in this set represents all the disagreements, just subset. For providing and running the BeadArrays as part of a Customer Service evaluation very different in probe design platforms... As mentioned, we expect higher expression levels to be affymetrix vs illumina positives or pose extra in... Available from the manufacturers ( Supplementary data or are available to you manufacturer... One was arbitrarily chosen ( 247 cases for Affymetrix microarray analysis, one can consider finer! Platforms in light of such considerations, most published gene expression level, GenBank! Uses multiple probes for a single transcript of a gene â1 and 1, apparently reflecting probes whose are... Genome ( i.e assayed in some cases of poor agreement across platforms eight arrays per,! Comparison between in situ synthesized oligonucleotide arrays, 2005 facilities for further study, limitations. Extra challenges in probe design ) on March 15, 2005 will ( definition. Donationsâ ( checks and money orders only ) to: we 're currently experiencing some in.