Title: S
1S M A S H
- Single
- Molecule
- Approach
- to
- Sequencing-by-Hybridization
2Bud Mishra
- Professor of Computer Science, Mathematics and
Cell Biology -
- Courant Institute, NYU School of Medicine, Tata
Institute of Fundamental Research, and Mt. Sinai
School of Medicine
3(No Transcript)
4Reading DNA Sequences
5How Does Nature Do It?
6Tools of the trade
- Where we collect three important tools from
biotechnology scissors, glues and copiers
7Scissors
- Type II Restriction Enzyme
- Biochemicals capable of cutting the
double-stranded DNA by breaking two -O-P-O
bridges on each backbone - Restriction Site
- Corresponds to specific short sequences EcoRI
GAATTC - Naturally occurring protein in bacteriaDefends
the bacterium from invading viral DNABacterium
produces another enzyme that methylates the
restriction sites of its own DNA
Tools of the Trade
8Glue
- DNA Ligase
- Cellular Enzyme Joins two strands of DNA
molecules by repairing phosphodiester bonds - T4 DNA Ligase (E. coli infected with
bacteriophage T4) - Hybridization
- Hydrogen bonding between two complementary single
stranded DNA fragments, or an RNA fragment and a
complementary single stranded DNA fragment
results in a double stranded DNA or a DNA-RNA
fragment
Tools of the Trade
9Copier
- DNA Amplification
- Main Ingredients Insert (the DNA segment to be
amplified), Vector (a cloning vector that
combines with an insert to create a replicon),
Host Organism (usually bacteria).
Tools of the Trade
10Copier
- PCR (Polymerase Chain Reaction)
- Main Ingredients Primers, Catalysts, Templates,
and the dNTPs.
Tools of the Trade
11Science by Stamp Collecting
- How biotechnology relates to the classical
coupon collectors problem
12Sir Ernest Rutherford
For Mikes sake, Soddy, dont call it
transmutation. Theyll have our heads off as
alchemists. Rutherford, winner of 1908 Nobel
prize for chemistry for cataloging alpha and beta
particles
- All science is either physics or stamp
collecting.
13Sanger Chemistry
14A Challenge
- Prize for DNA Decoding Aims to Fuel Innovation
- first team that completely decodes the DNA of 100
or more people in 10 days
15(No Transcript)
16Sequencing-by-SynthesisPyro-Sequencing
- BASE EXTENSION A single-stranded DNA fragment,
known as the template, is anchored to a surface
with the starting point of a complementary
strand, called the primer, attached to one of its
ends (a). When fluorescently tagged nucleotides
(dNTPs) and polymerase are exposed to the
template, a base complementary to the template
will be added to the primer strand (b). Remaining
polymerase and dNTPs are washed away, then laser
light excites the fluorescent tag, revealing the
identity of the newly incorporated nucleotide
(c). Its fluorescent tag is then stripped away,
and the process starts anew.
17The Middle Way
- Combining Two Extremes
- Indexing For each character b in the genome,
make a list of each position where it occurs. - Shotgunning For each long sentence in the
genome, select it with low probability
(o(lgn/n)), and then read it reasonably
accurately. - Where is the middle???
18The Middle Way
- Character Index
- A 1, 11,
- T 2, 3, 8, 9, 12
- C 4, 5, 9, 10, 13
- G 6, 7, 8, .
- Sentences w/o Index
- ATTCCGGG
- GGGCCATCGT
- CGTCATTCC
ATTCCGGGCCATC
ATTCCGGGCCATC
- Words w/ approx. Index
- ATTC 2..4
- TCGG 6..8
- GGGC 7..9
- GCCA 10..12
ATTCCGGGCCA
19Comparison
- Shotgun Assembly of Short Reads from
Sequencing-by-Synthesis/Ligation - Bottom-up
- Short high-quality reads without location
information - Shot-gun Assembler
- Must use maps or paired-reads
- Error Prone
- Genotypic Sequence
- Less Flexible
- SMASH Single Molecule Approach to
Sequencing-by-Hybridization - Top-down
- Shorter low-quality reads with location
information - Bayesian Assembler
- Haplotypic Sequence
- Flexible
- Gap-less Sequencing
- Large high-quality contigs
- Haplotype phasing
- Structural changes (e.g., chromosomal aberrations)
20Outline
- Physical Mapping Sequencing
- Map
- assign physical locations to important markers
(e.g., restriction sites or hybridization
probes). - Sequence
- align short sequence reads to the markers
(map-based sequence assembly) or - align long sequence reads to each other (shotgun
assembly) - Optical Mapping
- Sequencing
21Optical Mapping
- Where we map by watching the genome
22Probing a Single Molecule
- Standard ensemble measurements yield only average
values of a parameter for a large number of
identical copies of macromolecules. - Single molecule measurements provide rich set of
information - dependence of a parameter on its nano-environment
- statistical distribution
- dynamics (temporal effects)
23Optical Approaches
- Single DNA molecule on a surface can be explored
by nanometer scale with tunneling electrons,
forces from sharp tips or magnetic resonance - STM (scanning tunneling microscope)
- AFM (atomic force microscopy)
- MRFM (magnetic resonance force microscopy)
- Optical Approaches non-invasive, avoids
synchronization, need not be real-time.
24Optical Approaches are Inherently Noisy!
- Since many biological macromolecules are smaller
than the Raleigh limit, the optical approaches
involve attaching single fluorescent probes to
specific macromolecules. - Controlling Noise
- Magnitude of Stoke-shift
- Steric hinderance
- Absorption cross-section
- Point spread function (PSF)
- Image Processing
25Optical Mapping
26Optical Mapping
- Capture and immobilize whole genomes as massive
collections of single DNA molecules
Cells gently lysed to extract genomic DNA
DNA captured in parallel arrays of long single
DNA molecules using microfluidic device
Genomic DNA, captured as single DNA molecules
produced by random breakage of intact chromosomes
27Optical Mapping
2. Interrogate with restriction
endonucleases 3. Maintain order of restriction
fragments in each molecule
Digestion reveals 6-nucleotide cleavage sites as
gaps
28Optical Mapping
4. Determine size of fragments
29Optical Mapping
5. GENTIG Robust Bayesian Map Assembler to
make whole-genome restriction map
30Computational Analysis
Single DNA molecule on Optical Chip after
digestion, staining
- Image analysis software measures size and order
of restriction fragments - Overlapping single molecule maps are aligned to
produce a map assembly covering an entire
chromosome
31Map Assembly
Overlapping single molecule maps are aligned to
produce a map assembly covering an entire
chromosome
32Error Sources
- Sizing Error
- (Bernoulli labeling, absorption cross-section,
PSF) - Partial Digestion
- False Optical Sites
- Orientation
- Spurious molecules, Optical chimerism, Calibration
Image of restriction enzyme digested YAC clone
YAC clone 6H3, derived from human chromosome 11,
digested with the restriction endonuclease Eag I
and Mlu I, stained with a fluorochrome and imaged
by fluorescence microscopy.
33E. coli Shotgun Map
34Shotgun Mapping
- Large fragments of genomic DNA of length from 2Mb
to 12Mb are optically mapped - The resulting ordered restriction maps are
automatically contiged by Gentig - The consensus map computed by Gentig is free of
errors due to partial digestion, sizing error and
false cuts
35Taming the noise
- Where we examine many noise sources in optical
mapping
36Optical MappingInterplay between Biology and
Computation
37How Errors Complicate the Mapping
Correct Restriction Map
- Accuracy in Sizing
- False Cuts
- (governed by surface conditions, illuminations,
optic, imaging,) - Partial Digestion
Error Sources
- Orientation
- Spurious Data
- Crossing
- Breakage
- Missing Fragments
38Complexity
- Where we admit defeat with combinatorial
algorithmic approaches
39Complexity Issues
Various combinations of error sources lead to
NP-hard Problems
40SMRM(Single Molecule Restriction Map)
DRj
Dj
41SMRM(Single Molecule Restriction Map)
42Problem 2 (Sizing Error)
43Problem 2 is NP Complete
44Example
45Probabilistic Analysis
- Where we design the experiments to generate easy
instances of a difficult problem
46Sir Ernest Rutherford
- If your experiment needs statistics, you ought
to have done a better experiment.
47Combinatorial Structure
48Flips Flops
49Intuition
50Other Error Sources
51Discretization
52Sizing Error
53Prediction
The probability of successfully computing the
correct restriction map as a function of the
number of cuts in the map and number of molecules
used in creating the map
54Experimental Results
55Bayesian Methods
- Where we rely on statistical models of the error
sources to map correctly
56Bayesian Approach
57Robustness of Optical Mapping Algorithm
- BAC Clones with 6-cutters
- Average Clone size 160 Kb Average Fragment
Size 4 Kb, Average Number of Cutsites 40. - Parameters
- Digestion rate can be as low as 10
- Orientation of DNA need not be known.
- 40 foreign DNA
- 85 DNA partially broken
- Relative sizing error up to 30
- 30 spurious randomly located cuts
Algorithm Design and Analysis jointly with
T.S.Anantharaman
58Experimental Results
59Bayesian Inference
Pr(H D) Pr(D H) Pr(H)/ Pr(D)
60Bayesian Model
61Multiple Alignement
62Bayesian Optimization
Gradient search for good parameters
Local gradient optimization
63Y
- From a genes point of view, reshuffling is a
great restorative - The Y, in its solitary state disapproves of such
laxity. Apart from small parts near each tip
which line up with a shared section of the X, it
stands aloof from the great DNA swap. Its genes,
such as they are, remain in purdah as the
generations succeed. As a result, each Y is a
genetic republic, insulated from the outside
world. Like most closed societies it becomes both
selfish and wasteful. Every lineage evolves an
identity of its own which, quite often, collapses
under the weight of its own inborn weaknesses. - Celibacy has ruined mans chromosome.
- Steve Jones, Y The descent of Men, 2002.
64Mapping the DAZ locus on Y Chromosome
65GENomic conTIG
- Where we map large genomes
66Plasmodium falciparum
- Malaria Parasite
- Deadliest of all the human Malaria parasites
- P. falciparum
- P. vivax
- P.malariae
- P. ovale
- Responsible for 1.5-2.7 million deaths in 1997.
67Gentig MapsPlasmodium falciparum
- A. Gap-free consensus BamHI NheI maps for all
14 chromosomes. - B.BamHI map
- C. NheI map
- D.NheI map of Chromosome 3 displayed by ConVEx
68Gentig MapDeinococcus radiodurans
Nhe I map of D.radiodurans generated by Gentig
69Gentig MapE. coli
- Whole genome XhoI restriction map of E. coli O157
generated by Gentig software of Anantharaman
Mishra. - The outer circle represents an in silico XhoI
digest of the sequence. - The second outermost circle shows the consensus
optical map. - The inner circles represent the individual
molecule maps from which the consensus map was
generated.
70Complexity, Revisited
- Where we admit defeat again(?)
- Can Gentig assemble large Eukaryotic Genomes??
71GCP is NP-Complete
- Transformation from Hamiltonian Path Problem
restricted to cubic graphs.
Choose p 3/4 k M
72Comparing MapsEffect of Partial Digestion
- Parameters
- Partial digestion probability, p
- Relative sizing error, b
- Restriction fragments, n
- Overlap threshold ratio, q
- m n p Expected detected restriction
fragments. - Controlling False Negative
- K 5 np4 q/2 and r k1/p4, k1 ¼ 2
- If in fact the clones A and B overlap then we
will detect it with a probability, at least - (1-exp(-k1)) (1 exp(-n p4 q/8))
73Experiment Design
- Relation among the error parameters
- 3b n p /4 5 k 5 n p4 q/2
- ) p (3 b/2 q)1/3
- Parameter choice for shotgun-mapping. Make the
partial digestion probability rather high (close
to 1) or the relative sizing error as low for
instance by using a rare cutter.
74Contour Plot as a Function of Sizing Error
(x-axis) and Digestion Rate (y-axis)
- The calculation is for the human genome, G 3,300
mb. - The average molecule length 5 mb, with an
overlap of 1 mb - The average restriction fragment length 25 kb
- For a sizing error of 3 kb, the required
digestion rate is 94 - If the sizing error is reduced to 2 kb, the
required digestion rate drops to 88 - If the sizing error is reduced to 1 kb, the
required digestion rate drops to 80
75Sir Ernest Rutherford
- You should never bet against anything in science
at odds of more than about 1012 to 1.
76How does Gentig Work
- Gentig uses a purely Bayesian Approach.
- It models all the error processes in the prior.
- It initially starts with a conservative but fast
pairwise overlap configuration, computed
efficiently using Geometric Hashing. - It iteratively combines pairs of maps or map
contigs, while optimizing the likelihood score
subject to a constraint imposed by a
false-positive constraint. - It has special heuristics to handle non-local
errors.
77The problem Data error
- Miss restriction site rate
- Gaussian sizing var
- False restriction sites rate
- Small fragments missing rate
- Multiple chromosomes mixed together.
78Solution Bayesian probability maximization
H
f(HD1, L, Dm c f(H) f(D1, L, Dm H)
79Conditional Probability sum of alignments right
of I,J
H
80Example increase interval in H
81Sir Ernest Rutherford
- You should never bet against anything in science
at odds of more than about 1012 to 1.
82HAPTIG HAPlotypic conTIG
- HOOPLAS, HYPES HAPLOTYPES
- Replacing Gentig
- Faster More Accurate
83Single Molecule HapoltypingCandida Albicans
- The left end of chromsome-1 of the common fungus
Candida Albicans (being sequenced by Stanford). - You can clearly see 3 polymorphisms
- (A) Fragment 2 is of size 41.19kb (top) vs
38.73kb (bottom). - (B) The 3rd fragment of size 7.76kb is missing
from the top haplotype. - (C)The large fragment in the middle is of size
61.78kb vs 59.66kb.
84Goals
- Identify and phase polymorphism in data.
- Problem
- Similar to Gentig Problem
- if distributions of restriction site polymorphism
(due to SNPs) and restriction fragment length
polymorphism (due to indels) are known - How can one separate partial digestions from
restriction site polymorphism and sizing error
from restriction length polymorphism? - Solution
- Bayesian Analysis with Parameterized Priors
Fast Dynamic Programming Implementation.
85Comparison on 3.4 GHz CPU
Chr1 length Coverage USC time NYU time
12 Mb 20x 3431 secs 448 secs
12 Mb 100x 17052 secs 2818 secs
30 Mb 20x 28993 secs 1444 secs
30 Mb 100x 124194 secs 9308 secs
100 Mb 20x NA 25465 secs
100 Mb 50x NA 74025 secs
200 Mb 20x NA 88835 secs
200 Mb 50x NA 259024 secs
Note Haptig times based on 2.4 GHz CPU
86Comparison on 3.4 GHz CPU
20 X coverage
100 X coverage
Unlike the USC algorithm (by Waterman et al.),
Gentig/Haptig algorithms easily scale to human
genome sized problems. Gentig is routinely used
by OpGen and other labs to map.
87Other Interesting Applications
- Phylogeny
- Sequence Validation
- Haplotyping
- Sequencing
- Comparative Genomics
- Rearrangement events
- Hemizygous Deletions
- Epigenomics
- Characterizing cDNAs
- Expression Profiling
- Alternate Splicing
88Sequencing
89Joint Work with
- T.S. Anantharaman, NYU, NY.
- V. Demidov, UCLA, CA.
- A. Lim, NYU, NY.
- S. Paxia, NYU, NY.
- J. Reed, UCLA, CA.
- C. Cantor, Sequenom, Inc., CA
- J. Gimzewski, UCLA, CA.
- M. Teitell, UCLA, CA.
90OverviewTechnology
- How does it work?
- Optimal integration of several technologies based
on manipulation of single molecules on a surface.
91Goal
- Sequence a human size genome of about 6
Gbinclude both haplotypes. - Integrate three component technologies
- Optical Mapping to create Ordered Restriction
Maps with respect to one or more restriction
enzymes, - Hybridization of a pool of short nucleobase
probes (PNA or LNA oligomers) with Single Genome
dsDNA molecules on a surface, and - Efficient polynomial time algorithms to solve
localized versions of the PSBH (Positional
Sequencing by Hybridization) problems over the
whole genome.
92SMASH
- Genomic DNA is carefully extracted from small
number of cells of an organism (e.g., human) in
normal or diseased states. (Fig 1 shows a cancer
cell to be studied for its oncogeneomic
characterization.)
93SMASH
- LNA probes of length 6 8 nucleotides are
hybridized to dsDNA (double-stranded genomic DNA)
in a test tube (Fig 2) and the modified DNA is
stretched on a 1 x 1 chip that has microfluidic
channels manufactured on its surface. These
surfaces have been chemically treated to create a
positive charge.
DNA samples are prepared for analysis with LNA
probes and restriction enzymes.
94SMASH
- Since DNA is slightly negatively charged, it
adheres to the surface as it flows along these
channels and stretches out. Individual molecules
range in size from 0.3 3 million base pairs in
length. - Next, bright emitters are attached to the probes
on the surface and the molecules are imaged (Fig
3).
95SMASH
- A restriction enzyme1 is added to break the DNA
at specific sites. Since DNA molecules are under
slight tension, the cut fragments of DNA relax
like entropic springs, leaving small visible gaps
corresponding to the positions of the restriction
site (Fig 4). - 1. A restriction enzyme is a highly specific
molecular scissor that recognizes short
nucleotide sequences and cuts the DNA at only
those recognition sites.
96SMASH
- The DNA is then stained with a fluorogen (Fig 5)
and reimaged. The two images are combined to
create a composite image suggesting the locations
of a specific short word (e.g., probes) within
the context of a pattern of restriction sites.
97SMASH
- The intensity of the light emitted by the dye at
one frequency provides a measure of the length of
the DNA fragments. - The intensity of the light emitted by the
bright-emitters on probes provides an intensity
profile for locations of the probes. - Images of each DNA molecule are then converted
into ideograms, where the restriction sites are
represented by a tall rectangle and probe sites
by small circles (Fig 6).
98SMASH
- The steps above are repeated for all possible
probe compositions (modulo reverse
complementarity). - Sutta software then uses the data from all such
individual ideograms to create an assembly of the
haplotypic ordered restriction maps with
approximate probe locations superimposed on the
map.
99SMASH
- Local clusters of overlapping words are combined
by Suttas PSBH (positional sequencing by
hybridization) algorithm to overlay the inferred
haplotypic sequence on top of the restriction map
(Fig 7).
100Sir Ernest Rutherford
- We haven't the money, so we've got to think."
101Hybridization
102Thermal Stability of LNA
- LNA/DNA or LNA/RNA duplexes have increased
thermal stability compared with similar duplexes
formed by DNA or RNA. - The LNA modification has been shown to increase
the biological stability of nucleic acids. - Fully modified LNA oligonucleotides are resistant
towards most nucleases tested.
103Experiments with Probes
- To test the feasibility of hybridization of a
modified PNA, we first tested hybridization of
PNA probe to lambda DNA molecules. - We were able to measure the degree of
hybridization by digesting lambda DNA, after it
was hybridized with the PNA probe, with PmlI
restriction enzyme and run the sample on a 4
PAGE gel and found the degree of hybridization to
be greater then 90.
Bound
Unbound
104Probe Map
- Overlayed fluorescent images of labmda DNA
molecules using a FITC filter (white) and CY5
filter (red), showing the position of the probes
on the lambda DNA molecules.
105Final Probe Map
- We next combined all probe maps from around 20
image pairs using a Bayesian algorithm to compute
the most likely consensus map. - For one set of 20 image pairs, a total of 512 DNA
molecules with a total of 678 probes were
identified and combined into a consensus map with
2 probe locations at 14.8 and 52.4 of the DNA
length. - The orientation of the DNA cannot be determined
from optical maps. - Thus our result is in close agreement with the
correct map with probes at 50.2 and 85.7 known
from the sequence and the implied probe
hybridization rate of 42 is also quite good.
106Positional SBH Algorithm
107Algorithmic Problem
- The correct DNA sequence is s
- Computable
- Correct Haplotypic Restriction Maps (with respect
to multiple enzymes) - Positional Spectra For a position x, wx wcx,
the two k-mers at position x on the watson- and
crick-strands. - Data
- Map Errors Errors in sizing, false positives and
negatives in cleavage - Spectral Errors Errors in position and false
positives and negatives in hybridization - Compute a sequence t.
- How closely does t approximate s?
108Anomalies
- Irresolvable Ambiguities
- From assemblies based on 6bp probes
- Error Pattern s w sRC
- Correct Pattern s wRC sRC
- s tcgcc (any 5 bases)
- sRCggcga (Reverse compliment of X)
- w CCCCTAAC (any short sequence under 50bp)
- wRC GTTAGGGG (Reverse compliment of Y)
AssemblytcgccCCCCTAAC ggcga
Correct
tcgccGTTAGGGGggcga
109A Library of Anomalies
- Irresolvable Ambiguities Unavoidable Error
Patterns - Assuming assemblies with K-bp probes
- Most common s w sRC vs s wRC sRC
- Also common s w s t s vs. s t s w s
- Many more complicated patterns (but extremely
rare) - Here
- s any K-1 bp sequence
- w, t any short sequence under 50bp
- The probabilities of such patterns can be reduced
exponentially with gapped probes without
increasing the costs.
110Complexity
- Overcoming the complexity issues
- Spectrum of k-mer probes constraints on the
location of the k-mers in the sequence - However if the location constraints are in the
form of k contiguous locations then the
reconstruction problem is exponential in k,
rather than the sequence length m. - NP-Complete, if k 3
111Solution
- Complexity Issue
- a Experiment Design Powerful Heuristics
- Accuracy Issue
- a Combinatorial Design of Probes (Gapped Probes
using universal Bases)
112Gapped Probes
- Mixing solid bases with wild-card bases
- E.g., xxxxxx (9-mers) or xxxxxxxx (14
mers) - An inert base
- Universal In terms of its ability to form base
pairs with the other natural DNA/RNA bases. - Applications in primers and in probes for
hybridization - Examples
- The naturally occurring base hypoxanthine, as its
ribo- or 2'-deoxyribonucleoside - 2'-deoxyisoinosine
- 7-deaza-2'-deoxyinosine
- 2-aza-2'-deoxyinosine
1132'-Deoxyinosine derivatives
- 2'-Deoxyinosine derivatives can be used as
universal DNA analogues.
Loakes, D. Nucl. Acids Res. 2001 292437-2447
doi10.1093/nar/29.12.2437
114Other Chemistries
- The bases 3-nitropyrrole and 5-nitroindole have
also been incorporated into peptide nucleic acids
(PNA) - In PNADNA duplexes, it was found that the T_m
range was narrower than that found for
corresponding DNADNA duplexes, although
significantly higher than the corresponding
DNADNA duplex containing an AT base pair.
Loakes, D. Nucl. Acids Res. 2001 292437-2447
doi10.1093/nar/29.12.2437
115Approximation
- A special case of the PSBH problem,
- Separate constraints for each of the multiple
instances of each k-mer. - By focusing on a small window of about 2k-1bp, in
which most k-mers will occur only once, we hope
to approximately solve the standard PSBH problem - Here separate constraints for multiple instances
of each k-mer are not important. - Amortize over many instances
116Multiple PSBH problem
- Data set
- Probe maps for all K-mers (with gaps)
- We have multiple instances of each k-mer, (for
k6, about one every 4Kb on each strand of the
DNA in the sequence), - For each instance we can constrain the location
to within about 100 base pairs depending on the
optical resolution. - Beam Search Based Heuristics
117Simulation Results
- Probe Map Assumptions
- For single DNA molecules
- Probe location Standard Deviation 240 bases
- Data coverage per probe map 50x
- Probe hybridization rate 30, and
- false positive rate of 10 probes per megabase,
uniformly distributed. - Analytically estimation of the average error rate
in the probe consensus map - Probe location SD 60 bases
- False Positive rate lt 2.4
- False Negative rate lt 2.0.
118Simulation Results
- Using estimated error rates, randomly introduced
errors at the calculated rates into each of the
2080 simulated probe consensus maps (for the
above example). - Ran our sequence assembly algorithm.
- Aligned the sequence produced with the originally
assumed correct sequence using Smith-Waterman
alignment. - Counted the total number of single base errors
(mismatches deletions insertions). - Repeated this experiment until a total of 200,000
bases of sequence had been simulated and computed
the average error rate per 10,000 bases.
119Simulation Results
- We first tried un-gapped probes with 5,6,7 and 8
bases respectively and got error rates per 10,000
bases of 1674, 255, 39.6 and 3.7 bases
respectively. - Our machine ran out of memory trying to simulate
a 9 base ungapped probe, but from the trend it is
clear that the goal of 1 base error per 10,000
bases would most likely be achieved by a 9 base
ungapped probe.
120Results
UNGAPPED
GAPPED
1211000 Rupees Genome
22.67 US for 6 billion bases 135 billion US
for the entire human population
122Sir Ernest Rutherford
- I have become more and more impressed by the
power of the scientific method of extending our
knowledge of nature. - Experiment, directed by the imagination of either
an individual, or still better of a group of
individuals of varied mental outlook is able to
achieve results which far transcend the
imagination alone of the greatest natural
philosopher.
123Sir Ernest Rutherford
- Experiment without imagination, or imagination
without recourse to experiment, can accomplish
little. But for effective progress, a happy blend
of these powers is necessary
124the end