Using evSeq data¶

Successfully imported evSeq

In [2]:

import os
import glob

import numpy as np
import pandas as pd
import holoviews as hv
import ninetysix as ns

import evSeq.data_visualization as seqviz

Importing and viewing evSeq data¶

Reading in the sequencing files is easy! Just use pandas.read_csv() to look at the DataFrame. The function generate_platemaps() has been set up to work with the AminoAcids_Coupled_Max.csv file, so we'll look at this first.

In [3]:

# User input to relevant evSeqOutput folder
evseq_output_dir = '../data/example_notebook/SSM/'

In [4]:

# Read in the file
seq_file = 'AminoAcids_Coupled_Max.csv'
seq_files = os.path.join(evseq_output_dir, seq_file)
seq_df = pd.read_csv(seq_files)
seq_df

Out[4]:

	IndexPlate	Plate	Well	VariantCombo	SimpleCombo	VariantsFound	AlignmentFrequency	WellSeqDepth	VariantSequence	Flags
0	DI01	Lib1_105X	A01	?105R	R	1	0.986842	76	IIARTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
1	DI01	Lib1_105X	A02	?105D	D	1	1.000000	43	IIADTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
2	DI01	Lib1_105X	A03	?105W	W	1	0.981818	55	IIAWTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
3	DI01	Lib1_105X	A04	?105E	E	1	1.000000	57	IIAETGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
4	DI01	Lib1_105X	A05	?105P	P	1	0.970588	68	IIAPTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
...	...	...	...	...	...	...	...	...	...	...
757	DI08	Lib8_301X	H08	?301E	E	1	0.883459	266	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN
758	DI08	Lib8_301X	H09	?301P	P	1	0.990291	309	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN
759	DI08	Lib8_301X	H10	#DEAD#	#DEAD#	0	0.000000	0	#DEAD#	Too few usable reads
760	DI08	Lib8_301X	H11	G249D_?301L	DL	2	0.995370	216	CVSGGSNAAGIFYPFIDSGVKLIDVEAGGEGLETGKHAASLLKGKI...	NaN
761	DI08	Lib8_301X	H12	?301W	W	1	0.984456	193	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN

762 rows × 10 columns

As noted in Understanding the Outputs, this file contains information about the single most frequent variant in each sequenced well, as selected by the sequence with the highest alignment frequency.

When using NNN to specify variable codons, your VariantCombo output will return ? for the native amino acid (e.g., ?105V), since it does not know what the native codon is. The ? is hidden in the Platemap output, and other variability is given in the standard notation, like M97M above, since the codon in the reference sequence translated to M.

If you would like this in your Platemaps, you can use standard DataFrame functionality to add the parent amino acid when it is known:

In [5]:

# Create a dictionary of the native amino acids for each position
parent_dict = {
    # position: AA
    105: 'E',
    118: 'A',
    162: 'L',
    166: 'I',
    184: 'F',
    228: 'S',
    292: 'S',
    301: 'Y',
}

def set_parent(combo):
    """Parses the 'VariantCombo' string when it contains a ? for
    the native amino acid, given a parent_dict mapping; to be used
    as `df['VariantCombo'].apply(set_parent)` and reassigned
    back to the VariantCombo column.
    """
    # Get the mutations as a list
    muts = combo.split('_')
    
    # For each mutant
    for i, mut in enumerate(muts):
        
        # For the ? amino acids
        if mut[0] == '?':
            
            # Get the position given ?[postition]X
            position = int(mut[1:-1])

            # Get the parent amino acid
            parent_AA = parent_dict[position]

            # Replacethe '?' with this amino acid and replace in the list
            muts[i] = mut.replace('?', parent_AA)
    
    # Return in the same '_' joined format
    return '_'.join(muts)

# Apply the function
seq_df['VariantCombo'] = seq_df['VariantCombo'].apply(set_parent)

# Check
seq_df

Out[5]:

	IndexPlate	Plate	Well	VariantCombo	SimpleCombo	VariantsFound	AlignmentFrequency	WellSeqDepth	VariantSequence	Flags
0	DI01	Lib1_105X	A01	E105R	R	1	0.986842	76	IIARTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
1	DI01	Lib1_105X	A02	E105D	D	1	1.000000	43	IIADTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
2	DI01	Lib1_105X	A03	E105W	W	1	0.981818	55	IIAWTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
3	DI01	Lib1_105X	A04	E105E	E	1	1.000000	57	IIAETGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
4	DI01	Lib1_105X	A05	E105P	P	1	0.970588	68	IIAPTGAGQHGVATATAAALFGMECVIYMGEEDTIRQKLNVERMKL...	NaN
...	...	...	...	...	...	...	...	...	...	...
757	DI08	Lib8_301X	H08	Y301E	E	1	0.883459	266	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN
758	DI08	Lib8_301X	H09	Y301P	P	1	0.990291	309	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN
759	DI08	Lib8_301X	H10	#DEAD#	#DEAD#	0	0.000000	0	#DEAD#	Too few usable reads
760	DI08	Lib8_301X	H11	G249D_Y301L	DL	2	0.995370	216	CVSGGSNAAGIFYPFIDSGVKLIDVEAGGEGLETGKHAASLLKGKI...	NaN
761	DI08	Lib8_301X	H12	Y301W	W	1	0.984456	193	CVSGGSNAAGIFYPFIDSGVKLIGVEAGGEGLETGKHAASLLKGKI...	NaN

762 rows × 10 columns

The seqviz.generate_platemaps() function is used to create the Platemaps.html file at the end of every evSeq run. It can also be used to make a heatmap of the above updated data in a Jupyter notebook. This will output the same thing as the software does, but now within your Jupyter Notebook with any informative edits you have made.

Interpreting your heatmap:¶

• Borders are colored to show the alignment frequency. Ideally you have an alignment frequency of 0.90+. Below this, you may have a mixed well.
• Background of each well shows the log sequencing depth. We recommend at least 10 reads (which is color coded in white) before starting to trust the sequencing results. Bluer = better, redder = deader.
• You can hover over each well to find specific values for the mutations (for highly mutated variants these will not be shown as heatmap text and instead say 'N muts' to reduce clutter), sequencing depth, and alignment frequency.

In [6]:

single_site_sequencing = seqviz.generate_platemaps(seq_df)
single_site_sequencing

Out[6]:

Pairing sequence to function¶

Load in the data you have to pair with your sequencing data¶

We have also written some functions that help you pair your activity data with sequence data to examine single site-saturation libraries. To begin, you will want to read your activity data in as a DataFrame using pandas. You will want this DataFrame to include the columns Plate, Position, and Well in order to merge properly with the sequencing dataframe. You also want to avoid using other column names that are in the sequencing DataFrame since this could cause errors when merging.

In [7]:

# User input to functional data
func_file = '../data/example_notebook/SSM/201009_indole_rate_data.csv'

In [8]:

# Read in and check column names
func_df = pd.read_csv(func_file, index_col=0).rename(columns={'Plate_name': 'Plate'})
func_df

Out[8]:

	Position	Well	Rate (µM/s)	Type	Column	Row	Plate
0	105	A01	0.012341	Variant	1	A	Lib1_105X
1	105	A02	0.419586	Variant	2	A	Lib1_105X
2	105	A03	0.035158	Variant	3	A	Lib1_105X
3	105	A04	0.327427	Variant	4	A	Lib1_105X
4	105	A05	0.029853	Variant	5	A	Lib1_105X
...	...	...	...	...	...	...	...
763	301	H08	0.117740	Variant	8	H	Lib8_301X
764	301	H09	0.034030	Variant	9	H	Lib8_301X
765	301	H10	0.010226	Negative	10	H	Lib8_301X
766	301	H11	-0.011769	Variant	11	H	Lib8_301X
767	301	H12	-0.024868	Variant	12	H	Lib8_301X

768 rows × 7 columns

Parent control D10 in the 184 library had zero activity¶

Because this result is so much of an outlier based on known information, it is reasonable and prudent to discard it/relabel it as a Negative control.

In [9]:

# Remove 184-D10, which was a parent control but was dead
def set_negative(df_row):
    if (df_row['Position'] == 184) & (df_row['Well'] == 'D10'):
        df_row['Type'] = 'Negative'
        
    return df_row

# Apply the function
func_df = func_df.apply(set_negative, axis=1)

# Check
func_df[(func_df['Position'] == 184) & (func_df['Well'] == 'D10')]

Out[9]:

	Position	Well	Rate (µM/s)	Type	Column	Row	Plate
237	184	D10	-0.013732	Negative	10	D	Lib5_184X

Examine the number of "hits", using the thresholds of 1.2-fold and 2-fold improvements.¶

To demonstrate the cost-effectiveness of evSeq, we can look at how much it would cost to just sequence the improved variants in these eight plates.

In [10]:

normed = ns.normalize(
    func_df,
    value='Rate (µM/s)',
    to='Type=Parent',
    zero='Type=Negative',
    groupby='Plate',
    prefix='(normalized) '
)

# Examine the 1.2-fold hits from each plate
normed[normed['(normalized) Rate (µM/s)'] > 1.2].groupby('Position')['Well'].count()

Out[10]:

Position
105     1
118     2
162     3
184    36
228     8
Name: Well, dtype: int64

In [11]:

# Now the 2-fold
normed[normed['(normalized) Rate (µM/s)'] > 2].groupby('Position')['Well'].count()

Out[11]:

Position
162     1
184    11
Name: Well, dtype: int64

With only this function data, we would sequence somewhere around 12 to 50 samples then, one for each well that is a "hit".

For bulk samples you can do this with Sanger for \$3 (\\$36 to \$150), but for our normal services we do two reads (forward and reverse) at \~\\$4 each, meaning that this would actually cost \$96–\\$400.

Combine sequence and fitness data for use in exploring the data visually¶

Now you are ready to combine your data_df with the sequencing data.

Note: This is designed to work with single-site saturation mutagenesis sequencing data (needs a single Position value), and works best when the NNN codon was passed in explicitly.

In [12]:

# Function data is combined with the AminoAcids_Decouple_Max.csv file
seq_func_data = seqviz.combine_seq_func_data(func_df, evseq_output_dir)
seq_func_data

Out[12]:

	Position	Well	Rate (µM/s)	Type	Column	Row	Plate	IndexPlate	AA	AlignmentFrequency	WellSeqDepth	Flags	MutCount
0	105	A01	0.012341	Variant	1	A	Lib1_105X	DI01	R	0.987013	77.0	NaN	1.0
1	105	A02	0.419586	Variant	2	A	Lib1_105X	DI01	D	1.000000	43.0	NaN	1.0
2	105	A03	0.035158	Variant	3	A	Lib1_105X	DI01	W	0.981818	55.0	NaN	1.0
3	105	A04	0.327427	Variant	4	A	Lib1_105X	DI01	E	1.000000	57.0	NaN	1.0
4	105	A05	0.029853	Variant	5	A	Lib1_105X	DI01	P	0.971014	69.0	NaN	1.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...
763	301	H08	0.117740	Variant	8	H	Lib8_301X	DI08	E	0.882784	273.0	NaN	1.0
764	301	H09	0.034030	Variant	9	H	Lib8_301X	DI08	P	0.990536	317.0	NaN	1.0
765	301	H10	0.010226	Negative	10	H	Lib8_301X	NaN	NaN	NaN	NaN	NaN	NaN
766	301	H11	-0.011769	Variant	11	H	Lib8_301X	DI08	L	0.995516	223.0	NaN	2.0
767	301	H12	-0.024868	Variant	12	H	Lib8_301X	DI08	W	0.984925	199.0	NaN	1.0

768 rows × 13 columns

Analyzing single-site-saturation libraries¶

You can do sequence-function plotting a number of different ways.¶

Read the docstring to the plot_SSM_activities function with:

In [13]:

seqviz.plot_SSM_activities?

1. The most general use of this function.¶

Only paired sequence-function data is passed in and the name of the column that corresponds to the function. Here, red = more active (hotter) and blue = less active (cooler).

In [14]:

simple_seqfunc = seqviz.plot_SSM_activities(
    seq_func_data,
    value='Rate (µM/s)',
)

simple_seqfunc

Out[14]:

Notice that you can hover over individual points to find out more information about them, such as well, alignment frequency, and sequencing depth. Wells that did not pass the default quality control thresholds (>80% alignment, >10 quality reads, only 1 mutation found) are set in the Unknown column. (Also note that, for values with only single measurements these points are not present. Although it removes the ability to easily hover over them for more detail, we find that it is less misleading as it does not suggest that there are multiple overlaid measurements when others look at a static version of the plot. You may obtain this detail directly from the DataFrame as needed.)

2. Use known controls to understand the effect of your mutations¶

As described in the docstring, known is the name of a column that contains labels of the known control wells, and variant is the label used to denote which wells contain variants (not controls).

In [15]:

labeled_seqfunc = seqviz.plot_SSM_activities(
    seq_func_data, 
    value='Rate (µM/s)', 
    known='Type', 
    variant='Variant',
)

labeled_seqfunc

Out[15]:

3. Set the standard/parent to center the colormap¶

Finally, you can also pass the label of your standard (e.g., your parent control) from the known column which is set as the center of the colorbar (gray) and aligned to the left of the plot.

In [16]:

labeled_std_seqfunc = seqviz.plot_SSM_activities(
    seq_func_data, 
    value='Rate (µM/s)', 
    known='Type', 
    variant='Variant', 
    standard='Parent',
)

labeled_std_seqfunc

Out[16]:

evSeq for multisite libraries¶

You can also collect data for multisite libraries with evSeq!

In [17]:

# Load in the data and view
multi_seq_folder = '../data/example_notebook/Multisite_OutputCounts/'
multi_seq_files = os.path.join(multi_seq_folder, seq_file)
multi_seq_df = pd.read_csv(multi_seq_files)
multi_seq_df

Out[17]:

	IndexPlate	Plate	Well	VariantCombo	SimpleCombo	VariantsFound	AlignmentFrequency	WellSeqDepth	VariantSequence	Flags
0	DI01	Plate01	A01	?28L_?31M_?52W_?56H	LMWH	4	0.951258	636	ILATMGRLLFERYPETRSLFELPERWIHKHASA	NaN
1	DI01	Plate01	A02	?28M_?31F_?52S_?56V	MFSV	4	0.983478	1150	IMATFGRLLFERYPETRSLFELPERSIHKVASA	NaN
2	DI01	Plate01	A03	#DEAD#	#DEAD#	0	0.000000	9	#DEAD#	Too few usable reads
3	DI01	Plate01	A04	?28Q_?31G_?52E_?56D	QGED	4	0.977408	1682	IQATGGRLLFERYPETRSLFELPEREIHKDASA	NaN
4	DI01	Plate01	A05	?28I_?31I_?52S_?56L	IISL	4	0.978471	929	IIATIGRLLFERYPETRSLFELPERSIHKLASA	NaN
...	...	...	...	...	...	...	...	...	...	...
475	DI05	Plate05	H08	?28V_?31M_?52W_?56L	VMWL	4	0.990984	1220	IVATMGRLLFERYPETRSLFELPERWIHKLASA	NaN
476	DI05	Plate05	H09	?28H_?31E_?52T_?56L	HETL	4	0.983586	792	IHATEGRLLFERYPETRSLFELPERTIHKLASA	NaN
477	DI05	Plate05	H10	?28G_?31M_?52T_?56H	GMTH	4	0.957166	607	IGATMGRLLFERYPETRSLFELPERTIHKHASA	NaN
478	DI05	Plate05	H11	?28E_?31R_?52L_?56A	ERLA	4	0.986164	795	IEATRGRLLFERYPETRSLFELPERLIHKAASA	NaN
479	DI05	Plate05	H12	?28C_?31Q_?52R_?56F	CQRF	4	0.988535	785	ICATQGRLLFERYPETRSLFELPERRIHKFASA	NaN

480 rows × 10 columns

Again, we can use the same set_parent function from above to update our variant names with a new dictionary

In [18]:

parent_dict = {
    28: 'S',
    31: 'M',
    52: 'Q',
    56: 'L',
}

multi_seq_df['VariantCombo'] = multi_seq_df['VariantCombo'].apply(set_parent)

seqviz.generate_platemaps(multi_seq_df)

Out[18]:

Generating our visualizations¶

Below you will see code for generating the plots from the figures in the main text of the evSeq manuscript and GitHub pages.

To save as SVG files you may need to configure bokeh, selenium, and chrome/geckodriver.

In [19]:

from holoviews import opts

Figure 2b¶

In [20]:

seq_df = pd.read_csv(seq_files)

figure2b = seqviz.generate_platemaps(seq_df).opts(
    xticks=0,
    yticks=0,
    xaxis=None,
    yaxis=None,
    labelled=[],
    show_legend=False,
    title='',
    toolbar=None,
).opts({'HeatMap': dict(colorbar=False)})

# These are overlaid in Figure 2b
figure2b

Out[20]:

Preparing data for figure 3¶

In [21]:

# Add a normalized column to the df for plotting the seq/func data for single sites
orig_SSM_data = ns.Plate(seq_func_data, value_name='Rate (µM/s)')
orig_SSM_data = orig_SSM_data.normalize(
    value='Rate (µM/s)',
    to='Type=Parent',
    update_value=True,
    groupby='Plate',
    prefix='Normalized ',
)

# Convert to DataFrame and rename column
orig_SSM_data = orig_SSM_data.df.rename(columns={'Normalized Rate (µM/s)': 'Normalized Rate'})

# Clean the data for plotting platemaps of all data; copy first
cleaned_seq_func_data = seq_func_data.copy()

# Don't need Flags
cleaned_seq_func_data = cleaned_seq_func_data.drop(columns=['Flags'])

# Drop na columns
cleaned_seq_func_data = cleaned_seq_func_data.dropna()

# Take only wells that match a certain critera
cleaned_seq_func_data = cleaned_seq_func_data.loc[(
    (cleaned_seq_func_data['AlignmentFrequency'] > 0.8) &
    (cleaned_seq_func_data['WellSeqDepth'] > 10) &
    (cleaned_seq_func_data['MutCount'] <= 1)
)]
cleaned_seq_func_data['Position'] = cleaned_seq_func_data['Position'].astype('str', copy=False)

# Create a plate object with ninetysix and add normalized columns
SSM_data = ns.Plate(cleaned_seq_func_data, value_name='Rate (µM/s)')
SSM_data = SSM_data.normalize(
    value='Rate (µM/s)',
    to='Type=Parent',
    update_value=True,
    groupby='Plate',
    prefix='Normalized ',
)

# Convert to DataFrame and rename column
SSM_data = SSM_data.df.rename(columns={'Normalized Rate (µM/s)': 'Normalized Rate'})

# Show the data
SSM_data

Out[21]:

	Well	Row	Column	Position	Type	Plate	IndexPlate	AA	AlignmentFrequency	WellSeqDepth	MutCount	Rate (µM/s)	Normalized Rate
0	A01	A	1	105	Variant	Lib1_105X	DI01	R	0.987013	77.0	1.0	0.012341	0.044932
1	A01	A	1	162	Variant	Lib3_162X	DI03	I	0.990431	209.0	1.0	0.201484	0.606387
2	A01	A	1	184	Variant	Lib5_184X	DI05	N	0.978947	95.0	1.0	0.633893	1.610422
3	A01	A	1	228	Variant	Lib6_228X	DI06	N	0.980237	253.0	1.0	0.391901	1.365036
4	A01	A	1	118	Variant	Lib2_118X	DI02	V	0.850746	67.0	1.0	0.247484	0.582869
...	...	...	...	...	...	...	...	...	...	...	...	...	...
618	H12	H	12	184	Variant	Lib5_184X	DI05	V	0.968254	63.0	1.0	0.137598	0.349571
619	H12	H	12	292	Variant	Lib7_292X	DI07	T	0.982353	170.0	1.0	0.325779	0.895431
620	H12	H	12	118	Variant	Lib2_118X	DI02	K	0.921875	64.0	1.0	0.015398	0.036265
621	H12	H	12	166	Variant	Lib4_166X	DI04	S	0.919355	62.0	1.0	0.119664	0.339730
622	H12	H	12	301	Variant	Lib8_301X	DI08	W	0.984925	199.0	1.0	-0.024868	-0.050703

623 rows × 13 columns

Figure 3b¶

In [22]:

import colorcet as cc
count_cmap = cc.coolwarm[int(len(cc.coolwarm)/2)::1]

# set height and width for the two SSM heatmaps
n_rows=8
n_cols=20
height = 250
width = height * n_cols // n_rows

# Get counts of non-control wells (# of actual sequenced unknowns)
SSM_counts = SSM_data.loc[SSM_data['Type']=='Variant'].groupby(by=['Position','AA']).count()

# Rename the count column
SSM_counts = SSM_counts.rename(columns={'Normalized Rate': 'Counts'})

figure3b = hv.HeatMap(
    data=SSM_counts,
    kdims=['AA', 'Position'], 
    vdims=hv.Dimension('Counts', range=(0,14)),
).opts(
    height=height,
    width=width,
    cmap=count_cmap,
    xrotation=45,
    colorbar=True,
    colorbar_opts=dict(
        title='Counts',
    ),
    xlabel='Residue',
    invert_yaxis = True,
    tools=['hover']
)

figure3b

Out[22]:

Figure 3c¶

In [23]:

color_levels=ns.viz._center_colormap(SSM_data['Normalized Rate'], 1)

figure3c = hv.HeatMap(
    data=SSM_data.groupby(by=['Position','AA']).mean(), # keeps parent in the calculations
    kdims=['AA', 'Position'], 
    vdims=['Normalized Rate'],
).opts(
    height=height, 
    width=width,
    cmap='coolwarm',
    color_levels=color_levels,
    colorbar=True,
    colorbar_opts=dict(
        title='Normalized Rate',
    ),
    xrotation=45,
    xlabel='Residue',
    invert_yaxis=True,
    tools=['hover'],
)

figure3c

Out[23]:

Figure 3d¶

In [24]:

# Just a look at position 105, full SSM activity plot
lib105X_plot = seqviz.plot_SSM_activities(
    orig_SSM_data.loc[(orig_SSM_data['Plate']=='Lib1_105X')],
    value='Normalized Rate',
    known='Type',
    variant='Variant',
    standard='Parent',
)

# Just pull the histogram from the layout (index=1)
figure3d = lib105X_plot[lib105X_plot.keys()[1]]
figure3d = figure3d.opts(color='Counts', cmap=count_cmap, tools=['hover'])

figure3d

Out[24]:

Figure 3e¶

In [25]:

# Activity plot is index=0
figure3e = lib105X_plot[lib105X_plot.keys()[0]].opts(
    opts.Bars(
        xlabel='Residue',
        ylabel='Normalized Rate',
    ),
)

figure3e

Out[25]:

Rendering figures as SVGs¶

Below are a list of commands to save Holoviews plots in SVG format. Some notes:

1. You may run into an error about selenium and/or Chrome/GeckoDriver. The ways to fix them are described in the error messages and realtively easy to perform.
2. Holoviews plots can be rendered to a bokeh plot, which provides more fine-tuning control over the layout (see #3).
   1. For Holomaps (anything with toggle bars), each plot must be saved individually, or else only the first plot will be saved. This is done by iterating through the keys as if it were a dictionary as done for figure 2b.
3. Bokeh plot attributes such as axis line width and font size can be changed by accessing those attributes (again, see code for Figure 2b).
4. Once the bokeh plot is satisfactory (check with bokeh.io.show(plot)), you MUST change the output_backend attribute to 'svg', as performed in each of these examples. Otherwise export_svg will not export an infinite-resolution SVG.
5. From there, more fine-tuning can be done in your favorite SVG editor (e.g., Illustrator, Inkscape).

In [26]:

# from bokeh.io import export_svg
# svg_path=''

# # figure 2b
# for key in figure2b.keys():
    
#     plot=hv.render(figure2b[key].opts(
#         opts.HeatMap(
#             colorbar=False,    
#         ),
#         opts.Points(
#             line_join='round'
#         )
#     ))

#     plot.outline_line_color='black'
#     plot.outline_line_width=5
#     plot.outline_line_alpha=1
#     plot.output_backend = 'svg'
#     plot.min_border=0
#     filename=svg_path+'figure2b_'+str(key)+'.svg'
#     export_svg(plot, filename=filename)


# figure 3b
# plot=hv.render(figure3b)
# plot.output_backend = 'svg'
# filename=svg_path+'figure3b'+'.svg'
# export_svg(plot, filename=filename)

# # figure 3c
# plot=hv.render(figure3c)
# plot.output_backend = 'svg'
# filename=svg_path+'figure3c'+'.svg'
# export_svg(plot, filename=filename)

# # figure 3d
# plot=hv.render(figure3d)
# plot.output_backend = 'svg'
# filename=svg_path+'figure3d'+'.svg'
# export_svg(plot, filename=filename)

# # figure 3e
# plot=hv.render(figure3e)
# plot.output_backend = 'svg'
# filename=svg_path+'figure3e'+'.svg'
# export_svg(plot, filename=filename)

Submitting to Protαβank and other databases¶

In [27]:

def protabank_exchange(evSeq_func_data=None):
    """Converts data from combine_seq_func_data to an acceptable
    format for submission to protabank
    """
    return NotImplemented

protabank_exchange()

Out[27]:

NotImplemented

(yet!)

Miscellaneous visualizations¶

Below are some short bits of lightly commented code that you can use to further analyze your evSeq data.

Look at the distribution of read counts/wellseqdepths¶

In [28]:

seqviz.check_distributions(multi_seq_df)

Out[28]:

In [29]:

seqviz.check_distributions(seq_df)

Out[29]:

For the second set, with 8 plates, the violin chart shows us something important: the distributions of sequencing depth (especially the highest values) for Lib6-8 are significantly higher than for the first 5. Importantly, these were constructed with a different set on inner primers, suggesting that the amplification with these primers was better than for the others. (The libraries are normalized, but this is done based on UV measurements post-gel purification and is not always perfect.)

---

Next page: Creating barcode/index pairs.

Back to the main page.