HD28 .M414 Dewey Q Analysis of Path-Based Approaches to Genomic Physical Mapping by Alan Rimm-Kaufman James Orlin WP #3742-94 November 1994 revised December 1994 ;/,AS£AC!MjSf;TTS INSVlTUTE OF TECH^!OLOGY JAN 19 1995 LIBRARIES Analysis of path-based approaches to genomic physical mapping Alan Rimm-Kaufman and James B. Orlini ^ J. A. Rimm-Kaufman, Operations Research Center, MIT, Cambridge, B. Orlin, Sloan School of Management and MIT, Cambridge, MA. 02139, USA the Whitehead Institute MA. 02139, MIT Center USA for Genome Research, Abstract An integrated genomic physical map combines multiple sources of data to position landmarks and clones along a genome. strategies for constructing integrated "Path-based" maps employ paths overlapping clones to bridge intervals between genetically markers. There is a difficulty rate of clone overlap error result in analytical difficulty, many may false linkages method to in of mapped with path-based strategies: a small create numerous spurious paths and the physical map. We evaluate path-based approaches propose an in light and demonstrate the method on an integrated map human genome. of this of the genomic physical maps combine many Integrated and positional data structural the varieties of order clones and landmarks along to Genetically-mapped sequence tagged sites (STSs) can genome. provide a backbone for an integrated physical map, with overlapping yeast chromosomes (YACs) used artificial provide coverage of the to The overlaps between genetic intervals between adjacent markers. YACs may be detected through a variety of fingerprinting and hybridization data. In we which each in two STSs, STS genetically Sj and s:, and maps to locus Sj lies in y^ by some introduce first is assigned length k between genetic locus and STS k i YACs, i, (ii) include only 1 to locus content mapping, and one YACs YAC and j (iii) for (iv) and yj j, yj^.i correspond A path of defined as is yk such that maps s, consider a a genetic locus. to y2 y-\, We terminology. and genetic locus (Vi.yi+l) the data indicate that the of length including order to discuss the results of our analysis to integrated physical maps, data set means (i) Sj si lies in each step overlap. to traditional content mapping, while longer paths depend on the detection of overlaps. An integrated map connecting genetically close two loci A that contains the will loci assign a if It is an STS YAC interval a short path joining the YAC. YAC to known is overlaps can create short false paths between genetically-mapped STSs, YACs is to Paths serious concern with such path-based mapping strategies that falsely detected of there YAC y-| resulting in spurious assignment genetic intervals, and to spurious coverage of the genome. in random graph theory (1) that, in certain random structures, all paths of bounded length suffice to connect essentially attention through Separation," in phenomenon has This pairs of points. recently gained popular John Guare's award-winning which it play, asserted that any two people is can be connected through a path of at most six if Degrees in of the world acquaintances. same phenomenon Within an integrated genomic physical map, this could occur even "Six A the majority of the clone overlaps are valid. small rate of falsely declared .clone overlaps can lead to spurious coverage of large regions of the To address this concern, genome. we developed a methodology the probability that a shortest path joining two then applied this methodology generation' integrated physical describe our methodology in to the map loci is CEPH-Genethon of the to assess We correct. 'first- human genome (2). We the context of this integrated physical map. Briefly, the CEPH-Genethon screening the 33,000-clone STS methods, the first CEPH mega-YAC library YAC screened by two different content mapping and Alu-PCR probe hybridization. method, 2100 genetically-mapped STSs against the half physical mapping data involved library (with half the partially to (3) In were screened STSs screened completely and obtain 1-2 positives). In the second method, Alu-PCR products were prepared from 5300 individual YACs and were screened by hybridization against spotted Alu-PCR products from a subset of 25,000 cell lines. based In addition, 'fingerprinting' YACs and monochromosomai many YACs were subjected (4). to hybrid hybridization- A path is Two YACs defined as before. least one of the two YAC, is either c, and (ii) each or hybridized to a set of (For technical reasons, 50% YACs had (a) at similar loci i and lie j on the same was used as an Alu-PCR probe that to the chromosomes monochromosomal panel chromosome that included c. chromosomal assignments were not always could be assigned to a single chromosome, hybridized to multiple to yj (i) gave no signal when hybridized unique: if j A chromosomally allowable (4). defined as a path where chromosome, and genetic locus are determined to overlap or (b) the two fragment fingerprints restriction I YACs was used as an Alu-PCR probe and hybridized to the other path between genetic locus of length k 19% chromosomes, and 31% could not be assigned any chromosome.) The CEPH-Genethon be the set of all chromosomally allowable paths experimental analysis was advanced given length were correct. It allowed, the coverage of the defined (2) to a specified of Neither analytic nor length connecting loci within 10 cM. 1, 3, map was integrated physical suggest that most paths of a to was noted genome that as longer paths are increases. With paths of length 5 and 7 YACs, the strategy covered 11%, 30%, 70%, and 87%, respectively, of the total genetic length of the percentage coverage centiMorgans measure of lying is defined loci. (2) between connected coverage when coverage does not account genetic in For example, genome. as the proportion loci. This is restricted to valid paths, for ends YACs of contigs that (The of total a conservative since the extend beyond the creating a path between two genetically indistinguishable STSs do coverage on the genome by this definition.) We set out to evaluate the not provide incremental CEPH-Genethon paths using data from the February, 1994 CEPH-Genethon data release QUICKMAP the publically released computer package accompanies the CEPH-Genethon data. minimum distance between them - Figure (7). may be connected, as a that first and using (6) which constructed the length chromosomally allowable path connecting every pair on the same chromosome of loci We (5) shows the proportion 1 We loci. of loci and the width function of the path length between the of the genetic interval regardless of the genetic were interested in determining what fraction of these observed paths were valid and what A is to were spurious. fraction simple consider way loci to estimate the proportion of false connections separated by >50 cM and connected by short paths. YAC Such paths must surely be spurious inasmuch as the average length is 1 genome. Mb, corresponding The proportion of to only about 1 0.05, 4, 33, and 63%, respectively. In connections dominate at these distances. not loci pairs at distances 5-10, much higher than suggests that many of the human loci pairs of length 3, 1, particular, the dramatically for path lengths exceeding four, connected in such widely-separated connected by chromosomally allowable paths is cM curve rises indicating that The proportion 10-20, and 20-50 for loci pairs at distances > 50 cM. and 7 5, random of cM was This these paths are also spurious, and that such spurious paths span close as distant We loci loci same pairs at approximately the rate pairs. attempted to We estimate the proportions of valid paths. partitioned the pairs of loci into six groups on the basis of their inter-locus distance: the groups consisted 10-20, 20-50, and > 50 cM, respectively. range of loci at distance path between them, d, let invalid path between them. Figure 1 then, distribution We them, and loci is (ii) loci is A^-j of the six groups. (i) independent the frequency of spurious of the distance V^ and two points spurious path of length < in a. of length < a. in distance range Let S(j(a) = the distance range d are connected by a Let 7t(j(a) = the probability that two follows from the independence of \^ and V^j that quantity probability distance range d are connected by any path of length < in + Sd(a) joint Under A^j. d are connected by a valid path It between of the length of the shortest valid path. Let P(j(a) = the probability that two points points min(V(-j,l(-j). the cumulative probability these assumptions, one can estimate an empirical probability that = the length of the shortest spurious path joining two independent density for 5-10, 2-5, length of the shortest With these definitions, each 0-2, the length of the shortest valid made two assumptions: paths between two at denote the length of the shortest may be viewed as of A^j for next l^j loci For a randomly chosen pair A^ denote the V^ denote let path between them, and , let of - Pd(a)Sd(a). 7r(j(a) is Thus, Pd(a) = {n^{a) directly observable. - n^ia) = Pc|(a) Sd(a))/(1-Sd(a)). We assume a. that Scj(a) The is 8 independent by at least and so can be calculated from points separated of d, Therefore, Pc|(a) can be calculated. 50 cM. One can then estimate, for each distance group and each path length, the probability valid shortest path This is 7u<j(a) - among A^=a) & and there is. is - one at least is The denominator the probability there no path Under the independence equal to (Pd(a) that there Ajj=a)/P(A(-j=a). The numerator 71^(3-1). a-1. | the observed shortest paths (Figure 2). the probability P(Vjj=a path of length a most P(V(j=a a valid is (valid or invalid) of length at of V^j and With Pd(a-1))(1 -8^(3-1)). the numerator 1^, probability, this provide a corrected estimate of the observed proportion of pairs connected by at least one out the expected fraction of whose pairs is we loci by subtrscting valid path (Figure 3) loci is shortest paths are invalid. The results indicate that for two loci within 5 cM of one another and spsnned by shortest p3ths of < 3 YACs, the shortest P3ths are if likely to YACs the shortest paths are > 4 cM one include at least or if the two 3part, then the shortest paths are likely to Considering only shortest p3ths of 3t most 3 within 5 cM, 3nd using the CEPH/Genethon d3t3 cover coverage of the genome is definition of only 3bout far less The previous 3n3lysis W3S including both of loci be the other hand, are more th3n 5 invalid. YACs between cover3ge given 27% of the in loci (2), the hum3n genome. than indicated in This (2). C3rried out on the complete d3t3 set Alu-PCR connections 3nd conducted two 3ddition3l sets On valid path. fingerprint d3t3. 3n3lyses -- We one with Alu-PCR but no fingerprints, and one with fingerprint data but no Alu-PCR -- to investigate whether the coverage by valid paths might increase. amount coverage may possibly increase with the omission of valid a data source The many spurious the omitted data has provided if The connections and few valid ones. results of both of these in analyses were qualitatively the same, and both additional cases resulted first a decreased coverage of the in analysis. the analysis, involving In fingerprints, two for 4 YACs one or paths are within 2 the two cM On of the probability each is In one another and spanned by the other hand, invalid. cM that the shortest path apart, then the shortest is YACs between (without fingerprints) cover include the shortest paths are > that are loci and spanned by a shortest path other, 55% if likely to (For example, for two only shortest paths of at most 3 Alu-PCR data of are more than 5 loci be likely to cM to the Alu-PCR connections but no YACs, the shortest paths are valid path. if within 5 loci shortest paths of < 3 at least genome as compared 24% of Considering invalid.) loci within 5 of the the analysis including fingerprints data but no 4 YACs, cM, human genome. Alu-PCR connections, the fingerprint data generate longer valid paths: within 5 cM paths of < 5 of fingerprint-only valid one another and spanned by shortest YACs are path. likely to be covered by at least between loci within one valid While the fingerprint data support longer overall. 5 loci fingerprint-only paths, the relative paucity of these paths leads to lower genomic coverage cM Fingerprint-only paths of < 5 cover only 21% of the the YACs human genome. 10 These specific analyses address neither the quality of the CEPH/Genethon data nor calculations consider only the quality of shortest paths CEPH/Genethon data as generated from QUICKMAP. of A in the larger fraction paths might be obtained using alternative analysis reliable as well as additional data since added by strategies, Genethon These alternative path-based strategies. (8) to their February, 1994 release. noisy data, modifying the Identifying CEPH/Genethon considering longer length paths in and removing definition of a path, addition to the map paths, or employing a denser genetic CEPH and minimum length are possible alternatives more that might yield a larger fraction of veracious paths covering of the genome. can indicate the Further, while our analysis probability that there is a valid path, it does not provide a means Only additional distinguish valid paths from invalid paths. experiments can provide laboratory invalidation of outlined a general In method to a specific data for estimating set, we have the percentage of valid the presence of false connections leading to invalid paths. order to estimate the probability that a shortest path needs is in confirmation or a path. Aside from an application paths final subtract out the probability that to it is invalid. estimated by evaluating the percentage of distance connected by shortest paths extends to to of each length. is valid, This, loci in pairs This methodology other analyses as well, including analyses using alternative criteria definitions of for paths. determining YAC overlaps or alternative one turn, 1 1 12 and References Notes (1) B. Bollobas, Random (2) D. Cohen, Chumakov, and I. Graphs. (Academic Press, London, 1985). Weissenbach, Nature, 366, 698 J. (1993). Weissenbach (3) J. et Nature, 359, 794 (1992). al., 1059 (1992). (4) C. Bellanne-Chantelot et (5) These data included 6602 Alu-PCR probes and 3450 STSs. the STSs were genetically map the resulting genetic data employed in Rigault et P. (7) We Cell, 70, mapped al., contained 1266 an additional 1270 Alu-PCR probes added with described which in in this available from rigault@ceph.cephb.fr. QUICKMAP of the paper correspond fingerprint links, fingerprint links (8) links; to loci. we conducted rules described settings, in (1). three analyses: once using once using only ALU The CLONESPATH to the default turn correspond to the path creation only fingerprint links. package, a variety of options and parameters; the results For the "link-type" parameter, ALU and to that described in (1). generate chromosomally allowable paths between selected offers of These data are the same loci. employed CLONESPATH, one module CLONESPATH 2035 a specific chromosomal location; to CEPH/Genethon beyond the dataset by (6) (1), a!.. links, and once using figures correspond to using the results were qualitatively the ALU and same using ALU only. Varying the criteria used to declare criteria used alternative to overlaps or modifying the declare a genetic interval to be covered leads to A few examples path-based construction strategies. involving overlap detection probes YAC to link include two YACs, and fingerprint overlap to link (ii) (i) requiring two or requiring both an two YACs. Examples more ALU ALU probe and a involving the definition 13 of covering a genetic interval include disjoint of this We thank D. Cohen, to I. a specific region Chumakov, and J. (iv) more YAC- restricting P. L. Kruglyak, Rigault for helpful discussions manuscript. Supported in part by L. the use on the genome. Weissenbach for sharing valuable data with us and the scientific community at large. thank E. Lander, D. Page, and requiring two or paths to cover the genetic interval, and each Alu-PCR probe (9) (iii) Stein, D. Cohen, I. Chumakov, and comments on the NIH grant HG00098. We 14 Figure Figure by Legends 1. Observed cumulative proportion inter-loci distance and path length. constructed between Figure 2. all of connected loci pairs, Minimal paths were intra-chromosomal loci Estimated probability that there is shortest path spanning a inter-locus interval at pairs. one least among all valid the observed shortest paths spanning that interval, by inter-locus distance and path length. Figure 3. Estimated true proportion of connected inter-locus distance covered correctly spanning the is and path length. The loci probability pairs, an by interval the probability that at least one of the paths interval is valid. is 1 00%n Q- O O _l a 75% 0) ^-» CJ c O u 50% c O O Q. O 25% D 0) > Q) o (\J CO ^ LO Path Length Figure 1 to c o t1 o 60%n i- oO ^ 50%- 0) < 40%(0 Q- "o O 30%- T3 4-* u a> c c o 20%- CJ 1 0% o Q. O Path Length Figure 2 MIT LIBRARIES llll IIIIIIHIIII llllj lllllll ilillllM lllllll 3 9080 00922 4038 0-2 cM > .55 0.75- Q. T3 > 0) (/) o c (O 4-' CO 0.25- (D O Path Length Figure 3. Barcode ON NEXT TOUST PAGE Date Sue