Difference between revisions of "HMM"

Revision as of 05:50, 9 October 2015

We use HMMs to detect phosphatase domains and accessory domains. We benefit from the HMMs in public database such as Pfam, and the sequence alignments from public database such as CDD, SMART and COG, which are useful to build HMMs. We also build HMMs from scratch.

HMMs for Determining the Boundaries of Protein Phosphatase Domain

• AP: AP
• CC1 Fold
  • DSP
  • Myotubularin
  • OCA
  • PTEN
  • PTP
  • Paladin
  • Sac

• CC2 fold
  • LMWPTP
  • SSU72

• CC3/Rhodanese/CDC25

• HAD fold
  • EYA
  • FCP
  • NagD

• HP fold
• HP1
• HP2

• PHP_PHP

• PPM_PPM

• PPP fold
  • PAP
  • PPPc

• RTR1

HMMs for Finding Protein Phosphatase Domain with High Coverage

We have built a semi-redundant HMM profiles to detect phosphatase domains from biological sequences. We first used the HMMs from public databases, Pfam, SMART and SUPERFAMILY to search protein phosphatases we collected from the literature. We found 1) the HMMs from public databases such as Pfam and SMART can not find all human protein phosphatases, 2) some HMMs are redundant, - they captured exactly the same set of protein phosphatases. Thus, we 1) build in-house HMMs to capture the human protein phosphatases missed, 2) remove 100% redundant HMMs. You can download the HMMs.

HMMs of accessory domains

• PAP_NTD: Purple Acid Phosphatase, N-Terminal Domain (in-house)

• CDC25_NTD: CDC25, N-terminal domain (in-house)

• IQ: IQ profile built from SMART alignment

• PPIP5K_RimK

• STS_UBA

• MTMR5_C1: MTMR5, C1 domain (in-house)

• MTMR_GRAM: myotubularin, GRAM domain (in-house)

• SacN: Sac N-terminal domain (in-house)

• IPPc: Inositol Polyphosphate Phosphatase, catalytic domain homologues (built from SMART alignment)

• SAC9_NTD1: SAC9 N-terminal region 1. The domain is mostly found in plants, green algae, and amoebazoa.

• SAC9_NTD2: SAC9 N-terminal region 2. The domain is mostly found in plants, green algae, and amoebazoa.

• [[HMM_PD0147|WW] (built from SMART alignment)

• PP2C_C: Pfam HMM specifically matches with PPP1C but not other PPP subfamilies. The Pfam PP2C_C profile only match to PPP1C subfamily, but not other PPPc subfamilies. It overlaps with our in-house PPPc HMM profile.

Guidance for HMM building

We usually built the HMMs from PSI-BLAST hits.

To find the domain sequences for building a HMM, we PSI-BLASTed the domain sequence or the full sequence usually against protein NR/RefSeq/Swiss-Prot dataset via NCBI BLAST server. It sometime matters if you query the region that is supposed to contain the domain (based upon structure or any evidence) or the full sequence. The full sequence is often more sensitive to find weak hits to the domain. We recommended to download the files of Alignment, Search Strategies, and PssmWithParameters of PSI-BLAST result for reproductivity. The boundaries of the domain are determined by crystal structures, usually using the boundaries described in the papers reported the structures.

After several rounds of PSI-BLAST, we download the sequences of the aligned regions (not the complete sequences) from PSI-BLAST result. Because some sequences are redundant, which are not useful to build the HMM profile, we create the non-redundant sequence data set by using program CD-HIT (usually with sequence identity threshold as 70%, i.e. the parameter -c is set as 0.7).

We then carry out multiple sequence alignment (MSA) using programs such as MUSCLE, manually adjust the alignment usually by removing low-quality region in MSA editor such as JalView. We further inspect the distribution of sequence lengths in the MSA and remove the sequences which are shorter than most sequences in the MSA. How we remove the short sequences is dependent on the distribution and the MSA itself, which varies case by case.

We carry out MSA program and manually adjust the resulted MSA again after remove the short sequences. Then, we build HMM using program HMMBUILD. Depending on the format you use, you may need to convert the MSA into STOCKHOLM format before running HMMBUILD.

Note: The conserved regions (determined by sequence similarity) could be longer or shorter than the domains observed in crystal structures.