A step-by-step guide to ChIP-seq data analysis

Video Transcript

• 00:00 - 00:06: Hello. Welcome to Abcam’s webinar on a step-by-step guide to ChIP-seq data analysis.
• 00:07 - 00:12: Today’s principal speaker is Shi Chen, postdoctoral fellow in the lab of Dr. Sarah
• 00:12 - 00:19: Teichman at EBI and Sanger in Hinxton. Shi did his undergraduate studies at Health Science
• 00:19 - 00:25: Center in Peking University. He obtained his PhD in Professor Andy Sharrock’s lab
• 00:25 - 00:31: at the University of Manchester, where he investigated the DNA binding specificity of
• 00:31 - 00:38: several FOGHAT transcription factors using both traditional biochemical assays and state-of-the-art
• 00:38 - 00:44: genomic approaches. Shi’s work focuses on understanding how transcription factors
• 00:44 - 00:50: control the fate decision of mouse T helper cells by integrating PF binding data and gene
• 00:50 - 00:58: expression data. Joining Shi today will be Miriam Ferrer, project manager for cellular assays at
• 00:58 - 01:05: Abcam. Miriam completed her biology degree at the University of Barcelona and has a PhD from
• 01:05 - 01:13: Free University in Amsterdam. After completing her PhD, she joined the MRC Laboratory of Molecular
• 01:13 - 01:19: Biology in Cambridge. I will now hand over to Shi, who will start this webinar. Shi?
• 01:21 - 01:25: Thank you, Vicky, for the introduction, and thank you all for coming to the webinar.
• 01:26 - 01:32: So this webinar is more like a technical tutorial of how to perform ChIP-seq data analysis,
• 01:32 - 01:39: especially for wildlife biologists that do not have much programming and learning experience.
• 01:40 - 01:45: It is also useful for bioinformatics beginners who need to perform ChIP-seq data analysis in the future.
• 01:47 - 01:52: In this webinar, the starting point is that you’ve got your raw sequencing reads files,
• 01:53 - 01:58: most likely FASTQ files, and we will only cover sequencing data from the Illumina platform.
• 01:59 - 02:04: I will show you how to align or map the reads to a reference genome,
• 02:05 - 02:10: how to perform peak calling to identify enriched regions for the protein of interest,
• 02:12 - 02:19: and how to find DNA motifs enriched within the binding region, and how to assign binding sites
• 02:19 - 02:24: to genes and get enriched gene ontology terms to discover potential biological functions.
• 02:25 - 02:32: After this webinar, the audience should be able to get a general idea about how ChIP-seq datasets
• 02:32 - 02:39: look like and what you can expect from ChIP-seq data. Let me give you an outline about this
• 02:39 - 02:46: webinar first. We will cover all the basic ChIP-seq routines by using the software listed
• 02:46 - 02:54: in this slide. All the software here, they either have a web-based interface like Galaxy and GRID,
• 02:55 - 03:00: or they have a graphic user interface like FASTQC and SeqMiner.
• 03:01 - 03:04: So, no command-line experience is needed here.
• 03:06 - 03:14: The datasets I’m using here are FOCUS-A1 ChIP-seq experiments. FOCUS-A1 is a FOGHAT transcription
• 03:14 - 03:19: factor, and it is one of the first transcription factors that has been ChIPed genome-wide.
• 03:19 - 03:24: We will do a side-by-side comparison of a successful FOCUS-A1 ChIP-seq experiment
• 03:25 - 03:33: and a failed one. The successful FOCUS-A1 ChIP-seq data is from the MCF7 breast cancer
• 03:33 - 03:39: cell line published in Nature Genetics from Jason Carroll’s lab, shown on the left-hand side.
• 03:40 - 03:47: In this paper, the authors discover that FOCUS-A1 is absolutely required for estrogen
• 03:47 - 03:55: receptor DNA binding and activity in breast cancer cells. The failed one is from the LOVO
• 03:55 - 04:01: colon cancer cell line published in Cell from UC Tablet’s lab, shown on the right-hand side.
• 04:02 - 04:07: In this paper, the authors found that transcription factor binding tends to
• 04:07 - 04:13: occur in a highly clustered manner, and most clusters contain cohesins.
• 04:14 - 04:17: They performed hundreds of transcription factor ChIP-seq experiments,
• 04:18 - 04:24: and FOCUS-A1 is one of them. However, the FOCUS-A1 ChIP-seq dataset didn’t pass the
• 04:24 - 04:32: authors’ QC standards, so they labeled this as a failed experiment. This provides a perfect
• 04:32 - 04:38: opportunity for us to get an idea about what a failed experiment looks like, which is sometimes
• 04:38 - 04:45: even more important than other things. And thank the authors for making this data publicly available.
• 04:48 - 04:54: Now, we will start with QC of your sequencing reads using a software called FastQC.
• 04:55 - 05:00: It is developed at the Babraham Institute, and you can download it
• 05:00 - 05:03: from the website indicated on the top of the slide.
• 05:03 - 05:10: As you can see here, FastQC can be run interactively, so you are able to locate
• 05:10 - 05:15: your sequencing reads files and load them into FastQC by simply clicking your mouse.
• 05:17 - 05:25: Here, we have four files, two for FOCUS-A1 MCF7 dataset, and the other two for FOCUS-A1 LOVO
• 05:26 - 05:33: dataset. For each dataset, we got FOCUS-A1 ChIP sample and a control sample.
• 05:35 - 05:41: In the MCF7 dataset, the control sample is the input control, while in the LOVO dataset,
• 05:42 - 05:50: the control is the IgG control. Once you load the files by FastQC, you should see something
• 05:50 - 05:56: like this in the middle of the slide. It gives you many QC metrics of your sequencing reads.
• 05:58 - 06:03: The usage of FastQC is beyond the scope of this webinar, and we won’t cover all the metrics in
• 06:03 - 06:13: the program. Here, we will only focus on three metrics. The first one is basic statistics,
• 06:14 - 06:19: which contains some basic information about your sequencing reads, like read length and
• 06:19 - 06:25: number of total reads. Well, perhaps the most important piece of information here is the
• 06:25 - 06:33: encoding. It tells you how the quality score of the sequencing reads are encoded. The FOCUS-A1
• 06:33 - 06:42: MCF7 experiment is encoded by Sanger Illumina 1.9, and the FOCUS-A1 LOVO experiment is by
• 06:42 - 06:49: Illumina 1.5. We will need this information later, and I will come back to this later.
• 06:51 - 07:01: The second metric is read quality. In this graph, the X-axis shows all the individual bases from the
• 07:01 - 07:09: reads. For every base, the Y-axis is the box plot of the distribution of quality scores of all reads
• 07:09 - 07:17: across that particular base. The yellow boxes show from the 25th to 75th percentile,
• 07:18 - 07:23: and the red line indicates the median, and the blue line indicates the mean of the quality scores.
• 07:25 - 07:28: You really want the quality scores well above 20 to be good bases.
• 07:30 - 07:37: In the FOCUS-A1 LOVO data on the right, the quality drops roughly at the end of the reads.
• 07:37 - 07:42: In this case, you might want to trim the last or even five bases of your reads 07:42 - 07:48: before downstream analysis. In this webinar, we won’t do this just for simplicity.
• 07:50 - 07:55: The third metric we are looking at is sequence duplication level.
• 07:57 - 08:01: It tells you how unique your reads are within the library,
• 08:02 - 08:04: and this is a measurement for library complexity.
• 08:06 - 08:12: Due to the random nature of sequencing, what you expect to see is that most sequencing reads
• 08:12 - 08:21: only occur once, like the FOCUS-A1 MCF7 data on the left. However, in the FOCUS-A1 LOVO data on
• 08:21 - 08:28: the right, the majority of reads occur twice. They are also allowed to occur four times.
• 08:29 - 08:37: The overall unique sequencing read is shown above the graph here. The FOCUS-A1 MCF7 data
• 08:37 - 08:45: have more than 70% unique sequences, while the FOCUS-A1 LOVO data have only around 44%
• 08:45 - 08:50: unique sequencing reads. The total number of reads in the FOCUS-A1 LOVO data set
• 08:51 - 08:59: is more than 48 million. So, 44% indicates that there are about 21 million unique reads,
• 08:59 - 09:06: which is quite okay, actually. Now, it is worth mentioning that the FASTQC results are only
• 09:06 - 09:11: indications of the quality of your sequencing, but they cannot tell you whether your ChIP
• 09:11 - 09:19: experiments work or not. That’s what we are going to do next. Now, after getting a rough idea about
• 09:19 - 09:24: the sequencing quality, we are ready to align the reads to the reference genome.
• 09:25 - 09:33: To do this, we will use Bowtie in the Galaxy platform. Galaxy is a user-friendly online
• 09:33 - 09:39: system with many integrated bioinformatics tools. After registering and logging into the Galaxy
• 09:39 - 09:46: website, which is usegalaxy.org, you need to upload your FASTQ files to the Galaxy server.
• 09:47 - 09:54: Normally, you just click the upload file under the Get Data tab on the left. You can choose the
• 09:54 - 10:02: file from your computer to upload. However, note here, if the files are too large, they will fail
• 10:02 - 10:08: to upload via browser. So, it is better to upload your FASTQ files via Galaxy FTP.
• 10:08 - 10:13: Any FTP client will do the trick. Here, I’m using FileZilla as an example.
• 10:14 - 10:22: In the host, fill in the address of the Galaxy website, which is usegalaxy.org. The username
• 10:22 - 10:28: will be your email address when you registered your account on the Galaxy website. And then,
• 10:29 - 10:37: fill in the password. Simply click Quick Connect. You will see some log information below
• 10:37 - 10:45: indicating you have successfully connected to the Galaxy FTP. The left-hand side shows the
• 10:45 - 10:52: files on your local computer, and the right-hand side shows your files on the Galaxy FTP. They are
• 10:52 - 11:00: none at the moment. So now, locate your FASTQ files on your computer on the left,
• 11:01 - 11:06: then simply drag them to the right. The files will start to upload.
• 11:07 - 11:14: It is also recommended to zip your FASTQ files to save uploading time. The upload speed is,
• 11:14 - 11:19: of course, not as fast as your local network, but it is still quite decent.
• 11:20 - 11:24: When it is done, the files uploaded will be shown on the right.
• 11:26 - 11:32: Then, if you go to the Galaxy website, the files you just uploaded will be shown in the middle.
• 11:33 - 11:38: Now, we are ready to upload the files from the FTP to the Galaxy server.
• 11:40 - 11:46: Simply tick the file you are going to upload. Then, in the File Format drop-down menu,
• 11:46 - 11:54: choose the right file format. Our files are FASTQ files, and you can see here,
• 11:55 - 12:02: there are actually four different kinds of FASTQ files. The FASTQ-CS-Sanger format is for
• 12:02 - 12:06: color-based solid sequencing platforms, and we won’t cover this in this webinar.
• 12:08 - 12:14: So, the formats we should consider are FASTQ-Illumina, FASTQ-Sanger, and FASTQ-Solixa.
• 12:15 - 12:21: The main differences among these three FASTQ formats are the ways they encode sequencing quality.
• 12:23 - 12:27: The FASTQ-Sanger format uses Phred Score to present sequencing quality,
• 12:28 - 12:34: and it uses the ASCII 2 string of Phred Score plus 33 to encode the quality score.
• 12:35 - 12:41: And FASTQ files from Illumina pipeline 1.8 and after are encoded in this way.
• 12:41 - 12:47: I will show you later what exactly this means. If your sequencing experiments are done recently,
• 12:48 - 12:51: so most likely your FASTQ files will be in FASTQ-Sanger format.
• 12:53 - 13:01: The FASTQ-Illumina format uses Phred Score as well, but it encodes the quality by ASCII 2
• 13:01 - 13:10: string of Phred Score plus 64. FASTQ files from Illumina pipeline starting 1.3 and before
• 1.8 13:11 - 13:16: will be in this format. If you remember from the previous FASTQC results,
• 13:18 - 13:25: the FOCUS-A1 LOVO chip sample is from Illumina pipeline 1.5, so it should be in FASTQ-Illumina
• 13:25 - 13:33: format. And the rest of the three samples are all from Illumina pipeline 1.9, so they will be in
• 13:33 - 13:40: FASTQ-Sanger format. The FASTQ-Solixa uses a different system to encode the quality score.
• 13:41 - 13:46: And it is highly unlikely that your data files will be in this format unless you are analyzing
• 13:46 - 13:54: some very old Solixa data. More information about this can be found in this paper from NAR
• 13:54 - 13:57: and from the Wikipedia page at the bottom.
• 13:59 - 14:06: So now take the file and choose the right file format. Click the Execute button.
• 14:06 - 14:11: The file will start to upload from the FTP to the Galaxy server.
• 14:12 - 14:17: After they are done, they will appear on the right-hand side with green color.
• 14:19 - 14:25: If you click the little eye button here, you will see what is actually in the file in the middle.
• 14:26 - 14:35: We use the FOCUS-A1 MCF7 data as an example. Basically, the FASTQ file is just a simple text
• 14:35 - 14:43: file. Each sequencing read is represented by four lines of text. There are about 27 million reads in
• 14:43 - 14:52: this sample, so there are about 108 million lines of text in this file. Let’s look into details of
• 14:52 - 14:59: one read shown here on the top in the black box. The first line is the name of the read,
• 15:00 - 15:08: and it must start with an at character. The second line is the actual DNA sequence of the read.
• 15:09 - 15:13: The third line means nothing, but it must start with the plus character,
• 15:14 - 15:20: and only this plus character is required. The rest of the line is optional.
• 15:21 - 15:26: Since I got the data from SRA, this line is the same as the first line except for the beginning
• 15:26 - 15:33: character, and most FASTQ files only have a plus character at the third line of each read.
• 15:34 - 15:38: In this way, you reduce the file size and save storage space.
• 15:39 - 15:45: The fourth line is the ASCII string of quality score encoding at each space.
• 15:46 - 15:53: Here is an ASCII table that you can easily find via the internet, and let’s look into each space
• 15:53 - 16:01: of this read. So the first space of this read is A, and the quality string for this space is B.
• 16:02 - 16:10: By looking at the ASCII table, we can find that the ASCII code for string B is 66.
• 16:11 - 16:19: Since this is FASTQ Sanger format, the Phred score will be 66 minus 33, which is 33.
• 16:20 - 16:27: A Phred score of 33 indicates that the error probability of this space is 0.0005.
• 16:29 - 16:36: And the second space of this read is G, and the quality string for this space is the plus character.
• 16:36 - 16:44: Again, by looking at the ASCII table, we found that the ASCII code for the plus character is 64.
• 16:45 - 16:51: Then the Phred score for this space will be 64 minus 33, which is 31.
• 16:52 - 16:59: A Phred score of 31 indicates that the error probability for this space is 0.0008.
• 17:00 - 17:04: And you can check every space within this read by yourself in this way.
• 17:05 - 17:10: And that is basically how you read and interpret the data from FASTQ files.
• 17:10 - 17:14: Now, we will use Bowtie to align the sequencing reads to the genome.
• 17:14 - 17:19: But the Bowtie software in Galaxy only supports FASTQ Sanger files,
• 17:20 - 17:26: so we need to convert the FOCUS-A1 LOVO ChIP file from FASTQ Illumina to FASTQ Sanger.
• 17:27 - 17:34: To do this, on the left-hand side, click FASTQ Groomer under the NGS QC and Manipulation tab.
• 17:34 - 17:40: Then in the middle, choose the file you are going to convert, and specify the input format,
• 17:42 - 17:48: and hit Execute. After it’s done, it will appear on the right with green color.
• 17:49 - 17:55: And you can click the pencil button here to rename it to something meaningful, like this.
• 17:58 - 18:02: Finally, we are ready to align or map the sequencing reads to the genome.
• 18:02 - 18:08: To align or map the sequencing reads to the reference genome, we will be using the Bowtie
• 18:08 - 18:15: program. So on the left-hand side, click Map with Bowtie for Illumina under the NGS Mapping tab.
• 18:16 - 18:24: Then in the middle, select the reference genome. In this case, all samples from both experiments
• 18:24 - 18:30: are from human cells. So we choose HG19 here, which is not the latest,
• 18:30 - 18:37: but it is a stable version of the Human Genome Assembly. For the input file, choose the FASTQ
• 18:37 - 18:44: file you want to map. And in the Bowtie setting, choose Full Parameter List. Then you will get
• 18:44 - 18:52: many options. Now we will keep everything at default, except change the “-m option to 1”.
• 18:53 - 19:00: So what does this mean? To put it simply, for some reads, they can be mapped to multiple
• 19:00 - 19:05: locations within the genome. And you can handle this read differently according to your need.
• 19:06 - 19:13: But now, we put one here, which simply discards those reads. This is a good starting point for
• 19:13 - 19:20: beginners. And actually, many people are still doing this when they analyze DNA-centric sequencing
• 19:20 - 19:29: data, like ChIP-seq, DNA-seq, and APEX-seq. So you do the same for the other three files,
• 19:29 - 19:33: and after they are finished, they will appear on the right-hand side in green color.
• 19:34 - 19:38: As you can see here, I already renamed them to something meaningful.
• 19:41 - 19:47: Again, you can click the eye button to actually see what’s inside in the output alignment file.
• 19:48 - 19:55: The output alignment file from Bowtie is called SAM file. The SAM file is also a simple text file
• 19:56 - 20:02: with each line representing a read containing the location it is mapped into the genome
• 20:02 - 20:09: and other information. The details about the SAM format can be found in the PDF link shown at the
• 20:10 - 20:19: top. Now, after mapping, we are going to perform peak calling to identify FOXA1 binding site.
• 20:21 - 20:28: However, the SAM files at the moment contain both mapped and unmapped reads. We only want
• 20:28 - 20:32: the mapped reads for the peak calling, so we need to remove all the unmapped reads.
• 20:33 - 20:40: To do this, we will use SAM tools shown on the left-hand side. Just click filter SAM or BAM,
• 20:41 - 20:48: then in the middle, choose the file you want to filter, and choose filter on bitwise flag to yes.
• 20:50 - 20:56: Then you will get more options. In the skip alignment with any of these flag bits set,
• 20:57 - 21:04: take the read as unmapped. So in this way, SAM tools will remove all the unmapped reads.
• 21:05 - 21:10: After it is finished, the output will appear on the right in green color,
• 21:12 - 21:19: and I already named them to something meaningful. By default, the output from SAM tools are BAM
• 21:19 - 21:26: files. The BAM file is a binary file, not a text file, so you won’t be able to view the content.
• 21:27 - 21:33: You can simply think of the BAM file as just a compressed SAM file. It significantly reduces
• 21:33 - 21:43: the file size. Okay, now we are ready to perform peak calling. For this, we will be using a very
• 21:43 - 21:50: popular peak caller called MACS, developed in Shirley Liu’s lab. On the left-hand side, under the
• 21:50 - 21:59: NGS peak calling tab, click MACS to use it. In the middle, there are many options, but we will only
• 21:59 - 22:07: change a few of them. First, give it a meaningful name, then choose the ChIP file, which is the
• 22:07 - 22:15: FOCUS-A1 ChIP file, and the control file, which is the matching control experiment. Input control for
• 22:15 - 22:26: MCF7, IgG control for LOVO cells. The default genome size is 2.7 billion base pairs, and this
• 22:26 - 22:32: is for human cells. For different organisms, check the MACS website for the genome size,
• 22:33 - 22:37: then change the effective size to your sequencing read length.
• 22:39 - 22:47: Take the output files into interval files. We will see this later. Then, choose save wig
• 22:47 - 22:54: files so that you will get the binding signal files later for visualization. Then last,
• 22:54 - 22:59: choose do not build the shifting model, and 100 base pairs for arbitrary shift size.
• 22:59 - 23:06: So, why change these two options? As you know, for a regular ChIP-seq experiment,
• 23:06 - 23:12: the sequencing reads you get are from the ends of your fragment, which are not the actual TF
• 23:12 - 23:17: binding sites. The actual TF binding sites are somewhere in the middle of the fragment.
• 23:19 - 23:25: By default, MACS will estimate the fragment length and shifting the reads to the middle by half of
• 23:26 - 23:35: the fragment length to represent the actual TF binding. However, for some reason, sometimes it
• 23:35 - 23:42: cannot reliably estimate the fragment size. A common practice is to simply disable this
• 23:42 - 23:50: and ask MACS to shift reads to a certain distance. In this case, we choose 100 base pairs,
• 23:51 - 23:57: which is the default when you disable the model build, and it works quite well.
• 23:59 - 24:06: Now, click execute. After it is finished, you will get some outputs on the right-hand side
• 24:06 - 24:12: in green color, and the number of output files depends on your MACS setting.
• 24:12 - 24:17: Well, in this case, we have six output files for each experiment.
• 24:18 - 24:24: Now, let’s have a look at what each file is. The first file is the HTML report containing the
• 24:24 - 24:32: running log of MACS. The second and third files are WIG files, containing the signal intensity
• 24:32 - 24:40: across the genome. The treatment WIG file is your ChIP sample signal, and the control WIG file is
• 24:41 - 24:50: your control sample signal. Again, the WIG files are simple text files. We won’t look at the
• 24:50 - 24:58: negative peaks interval files here. The fifth file here is a peak interval file containing the
• 24:58 - 25:07: enriched region of your protein, in this case, FOXA1. The last file is a bed file for visualizing
• 25:07 - 25:14: peak positions in a genome browser, and both the interval and bed files
• 25:14 - 25:20: are simple text files. If you click the name of the interval file,
• 25:21 - 25:26: you can download this interval file to your local computer, and it can be opened
• 25:26 - 25:32: and easily handled and manipulated by any spreadsheet software like Excel.
• 25:33 - 25:38: This is probably one of the most important output files. It contains all the information
• 25:38 - 25:44: you need to know about the binding site, including the location and the statistics.
• 25:45 - 25:53: Well, at this stage, the peak calling is done. A common and tricky question that most beginners
• 25:53 - 25:59: have to face at this stage is that, how can I tell whether my experiment works or not?
• 25:59 - 26:04: Actually, there are many things you can check, but the following three things are the most
• 26:04 - 26:11: efficient ways of telling whether the ChIP experiment works or not. First, by looking
• 26:11 - 26:20: at the interval file. Second, by visual inspection of the binding signal. And third, by looking
• 26:20 - 26:26: at the enriched motif within the binding site. We will now go through this one by one.
• 26:29 - 26:37: We will start with examining the interval files. To do this, you sort first by FDR,
• 26:37 - 26:46: smallest to largest. Then, by fold enrichment, largest to smallest. The first thing we should
• 26:46 - 26:54: look at is the number of binding sites. If you choose an FDR cutoff at 1%, which is commonly
• 26:54 - 27:06: used, there are about 42,000 peaks left in the FOXA1 MCF7 data, but only 430 peaks left
• 27:07 - 27:14: in the FOXA1 LOVO data. Of course, the number of binding sites depends on the protein,
• 27:14 - 27:20: the peak caller, the cell line, and many other things. But typically, the number you should
• 27:20 - 27:28: expect is from around 1,000 to several thousand, or even several tens of thousands. As you can see
• 27:28 - 27:37: here, the FOXA1 LOVO dataset apparently does not satisfy the expectation. The second thing
• 27:37 - 27:44: we could look at is the range of the fold enrichment over a local background. The FOXA1
• 27:44 - 27:53: MCF7 data have a wide range, from more than 300-fold to about 4-fold, while the largest
• 27:53 - 28:00: FOXA1 LOVO peak has only around 20-fold enrichment, indicating the signal-to-noise
• 28:00 - 28:08: ratio in this data is really low. The third thing we could look at is the number of sequencing reads
• 28:08 - 28:17: within the peak region. Again, the FOXA1 MCF7 data have a much wider range than the FOXA1 LOVO
• 28:17 - 28:25: data. So these three metrics we just went through clearly suggest that there is something wrong with
• 28:25 - 28:33: FOXA1 LOVO datasets. And the next thing we could do to judge the quality of the data is to visualize
• 28:33 - 28:39: the binding signal, and we found this seems to be the most efficient way of telling whether a
• 28:39 - 28:48: ChIP-seq experiment works or not. So the WIG files generated by MACS are the signal files,
• 28:51 - 28:57: and you can visualize them directly in the genome browser. However, for large datasets,
• 28:57 - 29:03: it is highly recommended to convert the WIG files to bigWIG files before visualization.
• 29:05 - 29:12: The bigWIG format is an indexed binary format. When you visualize the bigWIG files, only the
• 29:12 - 29:18: portions of the files being displayed are transferred to the genome browser, which is much
• 29:18 - 29:26: faster than loading the WIG files. So to convert WIG files to bigWIG, on the left-hand side,
• 29:27 - 29:33: click the WIG to bigWIG under the Convert Format tab. Then in the middle,
• 29:34 - 29:39: choose the WIG file you are going to convert, and click Execute.
• 29:41 - 29:48: After it is done, change the file name to something meaningful, and click the name of the file.
• 29:49 - 29:58: You will see an option to display at UCSC main. Then click that. A new tab or window of UCSC
• 29:58 - 30:04: genome browser will appear. In the genome browser, to configure the visualization,
• 30:05 - 30:10: set the display mode to full so that it will display the histogram of the signal.
• 30:11 - 30:18: Set the vertical viewing range from 0 to a reasonable maximum value, like 100 or 200.
• 30:19 - 30:25: And set data viewing scaling to use vertical viewing range setting, so that it won’t
• 30:25 - 30:32: automatically rescale. You would be able to visualize the binding signal like shown here.
• 30:34 - 30:40: Now, what kind of patterns are we expecting? There are a few things you can look at at this
• 30:40 - 30:48: stage. You can have a look at whether the peak shape is normal, i.e., whether they are smooth
• 30:48 - 30:55: bell-curve shaped peaks or some strange spikes. And if you have some non-targeted genes,
• 30:55 - 31:01: can you find some binding peaks in the promoters or near the transcription start sites?
• 31:02 - 31:05: And finally, to get a bigger view of the binding pattern,
• 31:06 - 31:08: you can look at the whole chromosome view.
• 31:11 - 31:17: This is an example of signals of all four samples across the entire chromosome 12 of the human
• 31:17 - 31:27: genome. The two control files are almost flat as you would expect them to be. The FOXA1-MCF7
• 31:27 - 31:34: ChIP sample has many peaks, but the FOXA1-LOVO ChIP sample looks very much like its matching
• 31:34 - 31:42: control sample. There are only a few very small peaks. Now, this is a clear indication
• 31:42 - 31:48: that the FOXA1-LOVO ChIP experiment failed, or at least it is not optimal.
• 31:49 - 31:53: So, that’s how you can judge the quality of the data visually.
• 31:54 - 31:59: The next step is to identify enriched motifs within the binding sites.
• 31:59 - 32:07: We will use a program called MEME-chip to do this. In order to use MEME-chip, we need to extract the
• 32:07 - 32:14: DNA sequence from the peak coordinates to a FASTA file. Now, I need to clarify two terms before we
• 32:14 - 32:23: go any further. When MACS is used for peak calling, peak means the whole enriched region from start to
• 32:24 - 32:33: end, indicated by the black bar on the top left, while summit, indicated by the thin orange line,
• 32:33 - 32:41: means the highest parallax point within the peak region. It is supposed to be the exact TF binding
• 32:41 - 32:49: site. It is not a good idea to put too many sequences for the motif discovery, and a common
• 32:50 - 32:57: practice for TF motif discovery is to use the 100 base pair region centered on the summit as the
• 32:57 - 33:06: input for motif discovery. So, the column E in the MACS interval file is the distance of the summit
• 33:06 - 33:14: to the start position. So, to create the coordinates of peak summit plus minus 50 base pairs,
• 33:15 - 33:21: just create four new columns after the FDR column of the interval file.
• 33:22 - 33:30: The first new column is the same as column A, which is the chromosome name. The second column
• 33:30 - 33:41: will be column B plus column E minus 50, and the third column will be column B plus column E plus
• 33:41 - 33:51: 50. In the fourth column, give a unique name for each individual region like this, and you can do
• 33:51 - 34:00: this very easily using Excel. Now, save these four new columns as a tab-delimited text file.
• 34:00 - 34:09: Let’s call it FOXA1-Summit-100BP.txt. For simplicity, I will only choose the top 1,000
• 34:09 - 34:16: regions for motif discovery in the webinar. Now, we are going to extract the 100 base pair DNA
• 34:16 - 34:25: sequence of this region using the coordinates. To do this, first upload this text file to the
• 34:25 - 34:34: Galaxy server using a web browser. On the left-hand side, click Upload Data under the ‘Get Data’ tab.
• 34:34 - 34:42: Then, in the middle, choose the file format as ‘Interval’. Choose the text file you just saved on
• 34:42 - 34:54: your computer, and make sure the genome assembly is correct. After uploading, click ‘Extract Genomic DNA’
• 34:54 - 35:02: under the ‘Search Sequences’ tab. Choose the file you just uploaded, and simply click ‘Execute’.
• 35:04 - 35:10: When it is finished, you will see the DNA sequence within each 100 base pair region
• 35:10 - 35:16: in a FASTA format, and you can save this FASTA file to your local computer.
• 35:17 - 35:27: Let’s call it FOXA1-Summit-100BP.FASTA. Now, go to the MEME-ChIP website using the address at the
• 35:27 - 35:36: top of the slide. The options here are pretty much self-explanatory. In the Input, choose the FASTA
• 35:37 - 35:44: file you just downloaded. Then, enter your email address and give a job name.
• 35:45 - 35:51: The default setting is always a good starting point, so you just simply click Start Search.
• 35:52 - 35:56: Then, you will receive an email with a link to retrieve the results.
• 35:58 - 36:04: The results page will look like this. The motif found at the left-hand side
• 36:04 - 36:08: is a de novo motif that you enriched within your input region.
• 36:09 - 36:16: And the known or similar motif on the right-hand side indicates the similarity between this motif
• 36:16 - 36:22: with known TF motifs. Now, shown on the screen is the FOXA1-MCF7 data.
• 36:23 - 36:32: The top motif is a typical 4K motif with a very small E value. This is pretty much what you expect
• 36:32 - 36:36: from a successful experiment. There are also some other motifs returned,
• 36:37 - 36:44: indicating potential interaction with other transcription factors. And if the family motif
• 36:44 - 36:48: of the transcription factor you are checking is not enriched in the binding region,
• 36:49 - 36:56: like the FOXA1-LOVO dataset we are going to show in the next slide, then you need to put a
• 36:56 - 37:03: question mark on the data unless you have other strong evidence suggesting that the experiments
• 37:03 - 37:11: are working. Now, shown on the screen is a motif result from the FOXA1-LOVO dataset.
• 37:13 - 37:18: As you can see here, although the top two motifs look like the 4K motif,
• 37:19 - 37:26: the E values are very high. There are also two motifs with very low complexities,
• 37:27 - 37:33: which are just simple repeats. This motif result further confirms that the FOXA1-LOVO
• 37:33 - 37:40: ChIP experiment is not working. Now, we have finished the motif discovery step,
• 37:41 - 37:44: and we have a better impression about the quality of the data.
• 37:46 - 37:51: Another thing that most biologists are interested in is to assign the binding
• 37:51 - 37:57: sites to genes and perform gene ontology analysis to find potential biological functions
• 37:57 - 38:03: of transcription factors. To do this, we will be using a web tool called GRID,
• 38:04 - 38:09: developed in Gil B. Gerano’s lab. It is very straightforward to use.
• 38:10 - 38:14: Just go to the GRID website using the address shown on the top right.
• 38:14 - 38:22: Then in the genome assembly, choose the reference genome. In this case, it’s HG19.
• 38:23 - 38:26: Then choose the peak file from your computer as test region.
• 38:28 - 38:33: Since this is a ChIP-seq experiment, choose the whole genome as background.
• 38:35 - 38:41: If you click show settings, you can see how exactly GRID assigns the peaks to genes,
• 38:42 - 38:45: and you can change it as you want. But according to the paper,
• 38:45 - 38:50: the default setting seems to perform the best. So just click submit.
• 38:52 - 39:00: The output looks like this. The top contains some basic information about peak gene association,
• 39:01 - 39:06: and the bottom contains the enriched GO terms from specific categories,
• 39:06 - 39:10: like molecular function, biological process, and the cellular component.
• 39:11 - 39:18: As well as a lot of information from many other different databases. It is quite informative
• 39:18 - 39:24: to look at those, but we won’t go through them here. One thing that is particularly useful is
• 39:24 - 39:30: that if you click the job description on the top, this section will expand.
• 39:32 - 39:37: And click view all genomic region gene association. You will see two tables.
• 39:38 - 39:44: One gives you for each peak its associated genes, and the other gives you for each gene
• 39:45 - 39:50: its associated peaks, together with the distance between the peak and the gene.
• 39:51 - 39:58: And you can download them as a text file and save for future use. This is quite convenient.
• 39:59 - 40:02: Now we have finished the gene ontology analysis.
• 40:02 - 40:08: Last but not least, it is a good way of showing binding signal in HeatMap.
• 40:09 - 40:17: To do this, we will need the alignment files, which are the BAM files we created after the
• 40:17 - 40:24: mapping earlier. So click download dataset to save the BAM files to your local computer.
• 40:24 - 40:28: We also need a program called SIGMiner, developed in Lazaro Tora’s lab. 40:33 - 40:41: Just download SIGMiner from the address at the top. Run it according to your operating system.
• 40:43 - 40:45: Then you should see an interface like this.
• 40:45 - 40:54: First, to load data into it, in the load reference section, choose the peak file.
• 40:55 – 41:02: For example, the 100 base pair region text file we just created. Since this is a small file,
• 41:03 - 41:07: it is immediately loaded, and you can see the information in the upper center.
• 41:09 - 41:18: In the load aligned reads section, choose the BAM files just downloaded, and then click load files.
• 41:18 - 41:25: These two files will be loaded into memory, and this seems to be the time-consuming part
• 41:25 - 41:33: of all the analysis in SIGMiner. After it is done, the datasets will be shown in the middle.
• 41:35 - 41:42: When you click extract data, SIGMiner will plot the reads from each alignment file you just loaded
• 41:42 - 41:47: on top of your peak file and generate a heat map matrix.
• 41:49 - 41:56: After it is done, the result will show up on the right. If you right-click the result,
• 41:57 - 42:04: choose visualization of heat map. A heat map of the two files you loaded will be displayed.
• 42:04 - 42:08: You will also have a density plot at the bottom right.
• 42:10 - 42:15: The good thing about SIGMiner is that you can load multiple experiments at once,
• 42:15 - 42:22: and easily perform k-means clustering to find different binding patterns, and extract different
• 42:22 - 42:27: clusters for more detailed analysis. But that is beyond the scope of this webinar.
• Time stamps go out of wack here
• 42:32 - 42:38: So far, that’s all the basic ChIP-seq routines done without any programming and command line involved.
• 42:39 - 42:42: Hopefully, by running the workflow in the webinar on your own data by yourself,
• 42:42 - 42:48: you would quickly have a general idea about how ChIP-seq datasets look like,
• 42:48 - 43:03: and what we expect from ChIP-seq datasets. Here, I put some references, mainly from Nature Protocol, containing detailed step-by-step guidance, including troubleshooting tips of the
• 43:03 - 43:09: key aspects of ChIP-seq analysis. Some of them require some programming, or at least
• 43:09 - 43:13: command line experience. Well, it is highly recommended for wildlife biologists who are
• 43:13 - 43:21: doing genomic experiments to learn some basics about Linux usage, and perform the data analysis yourself
• 43:22- 43:34: . It makes you have a better understanding of the data itself. And once you’ve done that, all the things we just went through in this webinar can be done in just several simple commands.
• 43:34 - 43:43: Okay, this is the end of my presentation. Now, I will hand over to Miriam, who will introduce som
• 43:43 - 43:41: related products and protocols from Abcam.
• 43:46 - 43:49: And then, I will come back to answer questions. Thank you.
• 43:51 - 43:55: Thank you, Shi, for such an interesting and comprehensive presentation. 43:56 - 44:01: And I would like to take this opportunity to talk to you about some up-to-date products and resources.
• 44:02 - 44:08: We have just launched a high-sensitivity ChIP kit that has been specifically designed to be
• 44:08 - 44:15: used when there is a limited amount of sample available to perform ChIP. For example, when
• 44:15 - 44:22: working with patient material, transgenic myotissues, or stem cells. With the high-sensitivity
• 44:22 - 44:29: ChIP kit, you can obtain enrichment of your sequence of interest by using as little as 2,000
• 44:29 - 44:36: cells or half a milligram of tissue per reaction. You can get your result in only five hours,
• 44:36 - 44:42: so the experiment can be done in one working day. And the eluted DNA can be used straightaway in
• 44:42 - 44:51: sequencing, or microarray, or qPCR. Our high-sensitivity ChIP kit is part of our range of high-sensitivity
• 44:51 - 44:57: products designed for when there is a limited amount of starting material.
• 44:59 - 45:06: If you want to create a DNA library with a limited amount of DNA, our high-sensitivity DNA library
• 45:06 - 45:13: preparation kit for Illumina sequencing allows you to create a DNA library from only 0.2 nanograms
• 45:13 - 45:22: of DNA. And our ChIP-Seq high-sensitivity kit combines the benefits of the two kits I’ve just
• 45:22 - 45:29: mentioned. This kit is designed to carry out a successful ChIP-Seq starting directly from as
• 45:29 - 45:38: little as 10 to 5,000 cells. Bisulfite sequencing is the use of bisulfite treatment
• 45:38 - 45:44: to determine the pattern of methylation. To help you to take your methylation pattern studies
• 45:44 - 45:51: further, we now offer a couple of products to prepare post-bisulfite DNA libraries for Illumina.
• 45:52 - 45:58: With our post-bisulfite DNA library preparation kit, you can prepare a library using
• 45:59 - 46:05: pre-treated DNA in only five hours. If you haven’t treated your DNA,
• 46:05 - 46:14: then we recommend our bisulfite-seq high-sensitivity kit, which has all the reagents to perform the
• 46:14 - 46:21: bisulfite modification, followed immediately by a DNA library preparation step. The library is
• 46:21 - 46:28: modified and ready in only six hours. With no delay, I will pass the microphone back to
• 46:28 - 46:35: Shi, who is ready to answer some of your questions. Okay, thank you, Miriam.
• 46:37 - 46:43: The first question is that, is paired-end or single-end read data better? Well,
• 46:44 - 46:53: for most ChIP-seq data, single-end data is more prevalent. If you do
• 46:54 - 46:59: paired-end sequencing for a ChIP experiment, of course, the cost will be much more than the
• 47:00 - 47:06: single-end data. But if you look at the information you get from paired-end data,
• 47:06 - 47:12: in terms of ChIP-seq, you don’t get a lot compared to the single-end data.
• 47:12 - 47:18: So, in a cost-effective way, most people still do single-end data in terms of ChIP-seq.
• 47:21 - 47:29: And the second question is that, is input or IgG a better control? Actually, they are both okay
• 47:29 - 47:36: for ChIP-seq experiments as controls. Different people use different controls. For us, we prefer
• 47:36 - 47:40: input as a control because it provides much more complex libraries.
• 47:42 - 47:48: And the third question is that, is there a simple way to quantify changes in peak
• 47:48 - 47:54: states between two ChIP-seq experiments using the same antibody, using chromatin
• 47:54 - 47:59: from different cell types or conditions? Well, this is a very tricky and very good question.
• 48:00 - 48:09: Now, there are, in the ChIP-seq communities, a lot of people have put in a lot of effort
• 48:09 - 48:16: to developing the software to identify differentially bound peaks. But at the
• 48:16 - 48:22: moment, there’s no consensus about which method to use. There are already a lot of methods out
• 48:22 - 48:30: there. If you Google differentially bound ChIP-seq methods, then I’m sure you’ll get a lot.
• 48:30 - 48:36: And you can have a look at the math behind the methods and choose by yourself.
• 48:37 - 48:43: And another question is that, what are the differences between reads and peaks?
• 48:44 - 48:53: So, the reads are the raw sequencing reads that you get from the sequencing machine containing
• 48:53 - 48:59: the actual DNA sequences and the quality scores. Nowadays, they are usually between 50 base pairs
• 48:59 - 49:04: to 100 base pairs long. After mapping, the mapped reads contain the genomic locations where
• 49:04 - 49:11: the reads are aligned. So, the mapped reads will be the genomic coordinates of 50 base pairs to
• 49:11 - 49:17: 100 base pairs long, depending on your sequencing length. While peaks are the genomic
• 49:17 - 49:24: regions returned by the peak caller, they are the regions where the reads are aggregating or pile up
• 49:24 - 49:27: and pass certain statistical thresholds of your peak caller.
• 49:29 - 49:35: And then, this is the last question I will answer. So, the question is, if the number of
• 49:35 - 49:43: binding sites falls below 1,000, should I consider my experiment has failed? The short
• 49:43 - 49:52: answer to this question is that, no, you cannot tell whether your ChIP-seq experiments fail or not
• 49:52 - 49:56: solely based on the number of binding sites, because the number of binding sites depends
• 49:56 - 50:01: on many things, like the expression of the protein, the cell line, the peak caller,
• 50:02 - 50:06: and also the cutoff threshold used in the peak calling.
• 50:07 - 50:14: If you use the method in this webinar, the number we normally get from transcription factor ChIP-seq
• 50:14 - 50:20: is around 1,000 to several thousand. Well, if the number of binding sites falls below 1,000,
• 50:20 - 50:26: you need to put a question mark. However, you cannot simply judge whether the experiment
• 50:26 - 50:30: works or not solely based on the number of binding sites. You also need to check other
• 50:30 - 50:37: things, like motif results and visual inspection of your signal intensity.
• 50:39 - 50:43: Okay, so this is the last question. Now, I will hand over to Vicky,
• 50:43 - 50:45: and thank you all for attending this webinar.
• 50:48 - 50:51: Okay. Thank you, Shi and Miriam, for your presentations today.
• 50:52 - 50:57: We have received quite a lot of questions. Unfortunately, we were not able to answer
• 50:57 - 51:03: all of them. For those whose questions were not answered, our scientific support team will contact
• 51:03 - 51:08: you shortly with the response to your question. If you have any questions about what has been
• 51:08 - 51:14: discussed in this webinar, or have any technical inquiries, our scientific support team will be
• 51:14 - 51:20: very happy to help you, and they can be contacted at technical@abcam.com. 51:22 - 51:26: We hope you have found this webinar informative and useful to your work. 51:26 - 51:30: We look forward to welcoming you to another webinar in the future.
• 51:30 - 51:33: Thank you again for attending, and good luck with your research.